1

TRAINSCOPE

MOBILE VIDEO FOR TRANSIT SECURITY

SYSTEM DESIGN DOCUMENT

Revision 2.1 December 5, 2005

CMU-CyLab-05-002

Carnegie Mellon University CyLab - Visual Intelligence Studio, CIC-2218

4720 Forbes Ave. Pittsburgh, PA 15213

Contact: Dr. Yang Cai <ycai@cmu.edu>

This document is delivered as-is. The authors and Carnegie Mellon University are not responsible for any damages or losses. Please do not redistribute this document.

2

TABLE OF CONTENT

1 SYSTEM OVERVIEW 5

2 BACKGROUND 6

3 DIGITAL NETWORK CAMERAS 7 3.1 Camera architecture 7 3.2 Digital-Analog comparison 8 4 WIRELESS NETWORK 802.11G 9 4.1 Advantages of 802.11g 9 4.2 Wi-Fi security 10 5 DYNAMIC BANDWIDTH MANAGEMENT 11 5.1 Bandwidth computation 11 5.2 Network Bandwidth 12 5.3 TrainscopeViewer 14 6 MULTI-POINT ACCESSING 17 6.1 Network design 17 6.2 Signal Handover 18 6.3 Multiple bridge access points for multiple trains 21 7 DISTANCE ISSUES 23

8 VIDEO QUALITY 26

9 SCALABILITY 28 9.1 Storage 28 9.2 Network Hardware 28 9.3 Pittsburgh Technology Center test for throughput 34 9.4 Pittsburgh Technology Center test for handover 38 9.5 Pittsburgh Technology Center test for multiple bridges 41 9.6 Bombardier Transportation Test 45 ANNEXE 0 - GLOSSARY OF ACRONYMS 50

ANNEXE 1 – HARDWARE LIST SUMMARY 51

ANNEXE 1A - LINKSYS WET54GV2 DATASHEET 53

ANNEXE 1B – NETGEAR SWITCH 5 PORTS 55

ANNEXE 1C – AXIS 210 DIGITAL CAMERA 56

ANNEXE 1D – CISCO CATALYST 3500 SERIES XL 57

ANNEXE 1E - CISCO AIRONET 1200 SERIES 58

3

ANNEXE 2 : ANTENNAS SPECIFICATIONS 61

Tables

TABLE 1 : DIGITAL AND ANALOG COMPARISON TABLE 8

TABLE 2 : 802.11X COMPARISON TABLE 9

TABLE 3 : WIMAX VS. WI-FI 10

TABLE 4 : HTTP PARAMETERS FOR DYNAMIC MANAGEMENT 15

TABLE 5 : CONFIGURATION OF THE ACCESS POINT TO AVOID HANDOVERS 19

TABLE 6 : NETWORK RANGE FOR AIRONET 1200 ACCESS POINT (CISCO) 24

TABLE 7 : NETWORK CONFIGURATION FO SEVERAL NETWORK RANGE* 32

TABLE 8 : CONFIGURATION OF THE ACCESS POINTS (ACCESS POINT 1 & 2) 36

TABLE 9 : NETWORK MEASURES FOR THE PITTSBURGH TECHNOLOGY CENTER 37

TABLE 10 : RESULTS FOR THE PITTSBURGH TECHNOLOGY CENTER 38

TABLE 11 : CONFIGURATION OF THE ACCESS POINTS FOR A THRESHOLD EQUAL TO 11MBPS 39

TABLE 12 : RESULTS FOR THE PITTSBURGH TECHNOLOGY CENTER – HANDOVER TEST 40

TABLE 13 : RESULTS FOR THE PITTSBURGH TECHNOLOGY CENTER – MULTIPLE BRIDGES TEST1 43

TABLE 14 : RESULTS FOR THE PITTSBURGH TECHNOLOGY CENTER – MULTIPLE BRIDGES TEST2 44

TABLE 15 : RESULTS FOR BOMBARDIER TRANSPORTATION TEST 48

4

Figures

FIGURE 1 : TRANSIT AIRPORT SHUTTLE FROM BOMBARDIER TRANSPORTATION ............................................. 6 FIGURE 2 : NETWORK CAMERA ARCHITECTURE............................................................................................... 7 FIGURE 3 : IMAGE SIZE DIAGRAM.................................................................................................................. 11 FIGURE 4 : DYNAMIC BANDWIDTH SCENARIO................................................................................................ 12 FIGURE 5 : 38MBT/S USED FOR A STATIC MANAGEMENT (100%=100MBT/S) ................................................ 13 FIGURE 6 : 38MBT/S – 9MBT/S FOR A BASIC DYNAMIC BANDWIDTH MANAGEMENT (100%=100MBT/S) ...... 13 FIGURE 7 : EVOLVED DYNAMIC BANDWIDTH MANAGEMENT 22MBT/S – 8MBT/S.......................................... 13 FIGURE 8 : HTTP REQUEST TO CHECK THE STATUS OF THE CAMERAS OF THE NETWORK............................... 14 FIGURE 9 : HTTP REQUEST TO MODIFY RESOLUTION AND COMPRESSION...................................................... 14 FIGURE 10 : HTTP REQUEST TO MODIFY THE FRAME PER SECOND ................................................................ 14 FIGURE 11 : ONBOARD NETWORK DIAGRAM FOR TRAINET......................................................................... 17 FIGURE 12 : LAND NETWORK DIAGRAM FOR TRAINET................................................................................ 17 FIGURE 13 : NETWORK DESIGN FOR ACCESS POINT COLLABORATION............................................................ 18 FIGURE 14 : MULTIPLE BRIDGE ACCESS POINTS FOR MULTIPLE TRAINS ......................................................... 21 FIGURE 15 : NETWORK TRAFFIC VS. DISTANCE.............................................................................................. 23 FIGURE 16 : MAXIMUM BANDWIDTH FOR OUTDOOR CONDITIONS 26 MBTS/S................................................ 23 FIGURE 17 : LINKING QUALITY VS. DISTANCE FROM ACCESS POINTS ........................................................... 24 FIGURE 18 : COVERAGE OF THE OMNI DIRECTIONAL ANTENNA ..................................................................... 24 FIGURE 19 : IMAGE QUALITY FOR 640X480 WITH BACKLIGHT COMPENSATION ............................................. 27 FIGURE 20 : BANDWIDTH VS. NUMBER OF CAMERA PER VEHICLE AT 30 FPS FOR DIFFERENT RESOLUTION.... 29 FIGURE 21 : BANDWIDTH VS. NUMBER OF CAMERA PER VEHICLE AT 15 FPS FOR DIFFERENT RESOLUTION.... 30 FIGURE 22 : BANDWIDTH VS. NUMBER OF CAMERA PER VEHICLE AT 30 FPS FOR DIFFERENT COMPRESSION.. 31 FIGURE 23 : BANDWIDTH VS. NUMBER OF CAMERA PER VEHICLE AT 15 FPS FOR DIFFERENT COMPRESSION.. 31 FIGURE 24 : BANDWIDTH VS. NUMBER OF VEHICLE AT 15 FPS 320X240 30% ............................................... 32 FIGURE 25 : NUMBER OF BRIDGES VS. NUMBER OF VEHICLES (4 CAMERAS OR 6 CAMERAS) AT 30 FPS .......... 33 FIGURE 26 : NUMBER OF BRIDGES VS. NUMBER OF VEHICLES (4 CAMERAS OR 6 CAMERAS) AT 15 FPS .......... 34 FIGURE 27 : ONBOARD NETWORK DIAGRAM FOR PTC TEST ........................................................................ 34 FIGURE 28 : LAND NETWORK DIAGRAM FOR PTC TEST ............................................................................... 35 FIGURE 29 : CAMERA SET UP ......................................................................................................................... 36 FIGURE 30 : SATELLITE VIEW PITTSBURGH TECHNOLOGY CENTER TEST ...................................................... 37 FIGURE 31: LAND NETWORK DIAGRAM FOR PTC TEST................................................................................ 38 FIGURE 32 : SATELLITE VIEW PITTSBURGH TECHNOLOGY CENTER TEST ...................................................... 39 FIGURE 33 : BANDWIDTH VS. DISTANCE (THRESHOLDS 11 MBPS, 12 MBPS AND 18 MBPS) ........................... 40 FIGURE 34 : ONBOARD NETWORK FOR VEHICLE 1 ......................................................................................... 41 FIGURE 35 : ONBOARD NETWORK FOR VEHICLE 2 ......................................................................................... 41 FIGURE 36 : CAMERA SCREENSHOTS FOR MULTIPLE BRIDGES SCENARIO 1 .................................................... 42 FIGURE 37 : MULTIPLE BRIDGES WITH TWO VEHICLE FOLLOWING EACH OTHER............................................ 42 FIGURE 38 : CAMERA SCREENSHOTS FOR MULTIPLE BRIDGES SCENARIO 1 .................................................... 43 FIGURE 39 : MULTIPLE BRIDGES WITH TWO VEHICLES INTERCEPTING........................................................... 44 FIGURE 40 : BOMBARDIER TRANSPORTATION TEST TRACK SATELLITE VIEW................................................. 45 FIGURE 41 : SHUTTLE VEHICLE ON THE TEST TRACK ..................................................................................... 45 FIGURE 42 : ONBOARD NETWORK DIAGRAM FOR TRAINET......................................................................... 46 FIGURE 43 : LAND NETWORK DIAGRAM FOR TRAINET................................................................................ 46 FIGURE 44 : IMAGE RECORDED THROUGH THE WIRELESS NETWORK.............................................................. 47 FIGURE 45 : NETWORK TRAFFIC FOR BOMBARDIER TEST TRACK (AVERAGE 8 MBPS) (100% 100MBPS) ...... 47

5

1 System Overview The purpose of this project is to develop a concept-proofing real-time video surveillance technology for security of transit vehicles. Our objective is to investigate the Quality of Service (QoS), affordability and scalability of the mobile video system. Our approach is the open-system design, which is affordable and flexible. The system includes inexpensive off-the-shelf digital cameras on board and a mobile network connected to a land network through a wireless connection. The Dynamic Bandwidth Management software was developed to expand the bandwidth for the wireless transmission of streaming video. The project was carried out by Carnegie Mellon University, with partnership with Bombardier Transportation. Our tasks for this project include:

• Architecture of a multi-point wireless assessing network video system • Development of the dynamic bandwidth management model and software • Quality of Service (QoS) research for the relationship of video spatiotemporal

resolution, wireless bandwidth and distance, et al. • Scalability studies for multiple camera and multiple vehicle configurations • Applications of the mobile video, e.g. unattended baggage detection, etc.

From the field tests in Bombardier Transportation West Mifflin Test Track on July 27, 2005, the investigators found that the system design matches the criteria for a CCTV system. The video flow from the four cameras was transmitted wirelessly in real time. The dynamic bandwidth management used only 30% of the connection capacity for a single vehicle. The tests showed that the ‘line-of-sight’ of the wireless signal is more important than the distance, which was the original concern. This finding is useful in designing the wireless antenna and determining the locations of the wireless repeaters. The test also showed that a proper handover configuration can improve the throughput of the wireless network. The investigators have also developed additional systems based on the mobile video, such as the unattended baggage detection, etc. (in a separate Cylab Technical Report). These prototypes have been tested with realistic samples such as video from the inside of the airport shuttles (see attached appendix).

6

2 Background Terrorist attacks at transit systems have increased worldwide during the past decade. Forty percent of terrorist targets worldwide in 1998 were transportation targets, with a growing number against transit systems. The tragic event of September 11th, Madrid in 2004 and the bombings in the subways and buses of London in 2005 along with the continuous threat of terrorism prompted U.S. transit agencies to take stock of their security procedures and prepare for more life threatening situations.

Figure 1 : Transit airport shuttle from Bombardier Transportation

7

3 Digital Network Cameras 3.1 Camera architecture A network camera is fully bi-directional and also integrates with the rest of the system to a high degree in a distributed and scalable environment. This is a contrast to an analog camera that is a one-directional signal carrier which terminates at the DVR and operator level. A network camera communicates with several applications in parallel, to perform various tasks, such as detecting motion or sending different streams of video. A network camera can be described as a camera and computer combined into one unit. It connects directly to the network as any other network device. A network camera has its own IP address and built-in computing functions to handle network communication. Everything needed for viewing images over the network is built into the unit. A network camera has built-in software for a Web server, FTP server, FTP client and e-mail client. Other features include alarm input and relay output functions.

Figure 2 : Network camera architecture

The network camera's camera component captures the image -- which can be described as light of different wavelengths -- and transforms it into electrical signals. These signals are then are converted from analog to digital format and transferred into the computer function where the image is compressed and sent out over the network. The lens of the camera focuses the light onto the image sensor (CCD/CMOS). Before reaching the image sensor, the images pass through the optical filter, which removes any infrared (IR) light so that the "correct" colors will be displayed. (In day/night cameras, this IR-cut filter is removable to provide high quality black & white video during nighttime conditions.) The image sensor converts the image, which is composed of light information, into electrical signals. These electrical, digital signals are now in a format that can be compressed and transferred over networks.

8

3.2 Digital-Analog comparison

A digital network camera-based system

An analog camera-based system

Access As open or closed access as needed. Remote access to live images and remote administration of a network camera are possible from anywhere using a standard Web browser on any PC.

Closed circuit. No possibility for remote access.

Ease of use

- You can administer and view the images remotely using a standard Web browser on any PC.

- Images can be recorded on a hard disk, enabling easy search possibilities, easy storage and no image degradation or wear.

- The hard disk can be located at a remote location for security purposes.

- Remote administration or monitoring is not possible.

- Images must be stored on video tape cassettes, which require constant changing and lots of storage space. The quality of recorded images deteriorate over time.

- The video cassette recorder must be located near the camera. This could potentially enable unauthorized persons to have access to the video tape.

Quality Digital images do not lose quality in transmission or storage. A digital picture is created using Motion-JPEG. Once created, the image is free from degradation. Each frame within a video stream is sharp.

Image quality is lost when using long cables and the resolution of a magnetic tape is normally quite low. In addition, the quality of the recorded video deteriorates over time.

System requirements

Everything needed to stream live video over networks is included in the network camera. Simply connect the network camera to a network. View, record and administer from any networked PC (located anywhere).

Connection to a coaxial cable, to a multiplexer, to a video or time lapse recorder, and to a locally placed CRT (cathode ray tube) monitor.

Installation Simply connect a network camera to the nearest network connection and assign an IP address.

Attach a coax cable to each and every camera and connect to the multiplexer.

Cabling

One standard UTP (unshielded twisted pair) network cable can forward images from hundreds of network cameras simultaneously.

One cable can transport video signals from only one camera at a time. If you have two cameras, you have to have two cables. This often means large cable trunks filled with thick and sensitive cables that are connected to a locally placed control room.

Scalability Adding more network cameras to the system is easy. Very difficult. Each analog camera requires its

own cable. Image quality is lost when using long cables.

Cost

A high quality network cable typically costs 30 to 40 percent less than a standard coaxial cable.

A network cable can also support hundreds of network cameras and other devices.

An IP-based network infrastructure is often already in place, which means the cost is reduced to only that of the network camera(s).

Expensive coaxial cables. A classic RG59 75 Ohms coaxial cable typically costs 30 to 40 percent more than a high quality network cable.

In addition, more cable is required. Each analog camera requires its own cabling. High labor and maintenance demands, plus cost of the analog camera(s), video tape recorder and video tape cassettes.

Table 1 : Digital and analog comparison table

9

4 Wireless Network 802.11g 4.1 Advantages of 802.11g In 2002 and 2003, Wireless Local Area Network products supporting a new standard called 802.11g began to appear on the scene. 802.11g attempts to combine the best of both 802.11a and 802.11b. 802.11g supports bandwidth up to 54 Mbps, and it uses the 2.4 Ghz frequency for greater range. 802.11g is backwards compatible with 802.11b, meaning that 802.11g access points will work with 802.11b wireless network adapters and vice versa.

• Pros of 802.11g - fastest maximum speed; supports more simultaneous users; signal range is best and is not easily obstructed and not expensive.

Standard Maximum Bit Rate Fallback Rates Channels Provided Frequency Band Radio Technique

802.11 2 Mbps 1 Mbps 3 2.4 GHz FHSS or DSSS

802.11b 11 Mbps

5.5 Mbps

2 Mbps

1 Mbps

3 2.4 GHz DSSS

802.11a 54 Mbps

48 Mbps

36 Mbps

24 Mbps

18 Mbps

12 Mbps

6 Mbps

2 Mbps

12 5 GHz OFDM

802.11g 54 Mbps Same as 802.11a 3 2.4 GHz OFDM

Table 2 : 802.11x comparison table

One approach to increasing the physical transfer rates of wireless systems employs multiple antenna systems for both the transmitter and the receiver. This technology is referred to as multiple-input multiple-output (MIMO), or smart antenna systems. MIMO exploits the use of multiple signals transmitted into the wireless medium and multiple signals received from the wireless medium to improve wireless performance. MIMO can provide many benefits, all derived from the ability to process spatially different signals simultaneously. Two important benefits explored here are antenna diversity and spatial multiplexing. Using multiple antennas, MIMO technology offers the ability to coherently resolve information from multiple signal paths using spatially separated receive antennas. Multipath signals are the reflected signals arriving at the receiver some time after the original or line of sight (LOS) signal has been received. Multipath is typically perceived as interference degrading a receiver's ability to recover the intelligent information. MIMO enables the opportunity to spatially resolve multipath signals, providing diversity gain that contributes to a receiver's ability to recover the

10

intelligent information. While WiMax has a wider range but it’s multipoint devices are not as mature as 802.11g.

WiMax (802.16a)

Wi-Fi (802.11g)

Primary development Broadband wireless access Wireless LAN Frequency Band Licensed/Unlicensed

2G to 11 GHz 2.4 GHz ISM (g)

Channel bandwidth Adjustable 1.25 M to 20 MHz

20 MHz

Bandwidth efficiency < 5bps/Hz < 2.7 bps/Hz Half/Full duplex Full Half

Modulation BPSK, QPSK 16-,64-, 256-QAM

BPSK, QPSK 16-,64-QAM

FEC Convolutional Code Reed-Solomon

Convolutional Code

Encryption Mandatory – 3DES Optional AES

Optional AES

Access Protocol - Best Effort - Data Priority - Consistent Delay

Request/Grant Yes Yes Yes

CSMA/CA Yes

802.11e WME 802.11e WSM

Radio Technology OFDM (256 channels) OFDM (64 channels)

Table 3 : WiMax vs. Wi-Fi

4.2 Wi-Fi security

In Wi-Fi, encryption is optional, and three different techniques have been defined:

Wired Equivalent Privacy (WEP): An RC4-based 40- or 104-bit encryption with a static key.

Wi-Fi Protected Access (WPA): A new standard from the Wi-Fi Alliance that uses the 40- or 104-bit WEP key, but changes the key on each packet to thwart key- crackers. That changing key functionality is called the Temporal Key Integrity Protocol (TKIP).

IEEE 802.11i/WPA2: The IEEE is finalizing the 802.11i standard, which will be based on a far more robust encryption technique called the Advanced Encryption Standard. The Wi-Fi Alliance will designate products that comply with the 802.11i standard as WPA2. However, implementing 802.11i will typically require a hardware upgrade, so while the standard should be completed in mid-2004, it might be some time before it is widely deployed.

11

5 Dynamic Bandwidth Management 5.1 Bandwidth computation For wireless network the bandwidth is the limiting factor and displaying more than 10 video with high resolution, high frame rate and high quality is currently a big issue. Typically, for a set up of four cameras capturing 640*480 video at 30 fps with a compression rate of 10%, the bandwidth required is 36.4 Mbit/sec. This is the reason why a system with more than 12 cameras overtakes the capacity of traditional Ethernet cable (100 Mbits/sec). In practice, a video camera displayed on the small screen with a high resolution does not provide a significant improvement for the viewer and is a waste of bandwidth. The correlation between image size, compression and resolution level is illustrated schematically in Fig. 3. The size of the picture from the motion JPEG flow increases with the resolution and the quality (less compression).

1030

5070

90160*120

320*240

640*4800

10

20

30

40

Image Size (Ko)

Compression Rate (%)Image

resolution

Image Size function of compression and resolution

30-4020-3010-200-10

Figure 3 : Image size diagram

A model of this function is created in order to optimize the link between human perception and image size. We obtained three equations (1)(2)(3) to compute the bandwidth. The exponential equation (2) is linked to the nature of JPEG compression. JPEG is designed for compressing full-color or gray-scale images of natural, real-world scenes. JPEG removes redundant information from images which is why equation (2) is based on the exponential: the non-redundant information could not be totally removed and remains as the lowest size of the picture. The image resolution I in pixels and the compression rate percentage CR are the main parameters used to compute the image size S in pixels and the resulting bandwidth B (Mbits per second). First, the non-redundant information RI can be approximated by equation (1).

12

7.41*410*3.44 +−= IRI (1)

40.0)(* −= CRRIS (2)

SCFPSCNSB ***= (3)

The constant SC represents the complexity of a scene. This constant is in most instances equal to 8. It is used along with the number of cameras CN and the frames per seconds fps for each camera to carefully estimate the bandwidth.

5.2 Network Bandwidth To demonstrate the performance of our system, we compare our CMU-system to existing systems. We have established some real scenarios to analyze the bandwidth adaptation when switching between different views. First, the user is watching a display with 4 equal screens. The screen selected is displayed in a full screen mode for 10 seconds. Then the user is returned to the four screen view. This scenario (Fig. 4) is tested with default static parameters: resolution (640x480) and compression (10%). For the dynamic management, these two parameters will change according to the human behavior. As explained previously, the number of frames per second is maintained at 30 to guarantee a full movement perception.

Figure 4 : Dynamic bandwidth scenario

The first comparative test is based on static bandwidth management. The default resolution (640x480) and compression (10%) are selected permanently. These parameters are chosen in order to obtain high definition video when the user selects a screen in a full screen mode. The results (Fig. 5) were predictable. Because none of the parameters evolve, the bandwidth does not change and 38Mbt/s is used permanently to display 4 cameras in high definition.

13

Figure 5 : 38Mbt/s used for a static management (100%=100Mbt/s)

The second test is based on basic dynamic management of the bandwidth. When the user switches to the full screen mode, the bit flow from the three other video cameras is disabled. It results a rectangle signal (Fig. 6) with the high level corresponding to the four screen mode (38Mbt/s) and the low level to the full screen mode (9Mbts). However, the bandwidth used globally is still too large to support over 12 video cameras.

Figure 6 : 38Mbt/s – 9Mbt/s for a basic dynamic bandwidth management (100%=100Mbt/s)

In the third trial, we develop a full dynamic bandwidth management system in order to adjust the human bandwidth to the network bandwidth. When the watcher switches to a 4 screens display, the image resolution becomes smaller so that the image size diminishes according to the equation (3). As explained previously, the viewer does not notice a change for over a certain resolution. This is why the low resolution fits perfectly the small display. Globally the bit rate for the four screen display decreases from 38Mbt/s to 22Mbt/s (Fig. 7). The gain is approximately 43%.

Figure 7 : Evolved dynamic bandwidth management 22Mbt/s – 8Mbt/s

14

This system avoids wasting bandwidth without providing a significant improvement for the viewer. This method has been implemented in “TrainscopeViewer”, a software created by the Visual Intelligence Studio group to manage the bandwidth for wireless network. 5.3 TrainscopeViewer 5.3.1 HTTP request TrainscopeViewer is an application developed by Visual Intelligence Studio of Carnegie Mellon University. It manages up to n IP cameras at the same time (tested with 16) and provides up to 4 different ways to display the cameras (cf. TrainscopeViewer guide). The software is coded in C++ and uses the Visual .Net compiler. As said previously, an IP camera is a digital camera and therefore is bidirectional. The video flow can be managed while sending HTTP requests to the camera. The camera will adapt the flow according to the requests received dynamically. TrainscopeViewer manage all the cameras available at the same time, sending requests through the HTTP port. These requests are sent by the user who is switching between the different views. http://myserver/cgi/view/videostatus.cgi?status=1,2,3,4? HTTP/1.0 200 OK\r\n Content-Type: text/plain\r\n \r\n Video 1 = video Video 2 = no video Video 3 = no video Video 4 = video

Figure 8 : HTTP request to check the status of the cameras of the network http://myserver/mjpg/video.cgi?resolution=320x240&camera=1&compression=25

Figure 9 : HTTP request to modify resolution and compression

http://myserver/cgi/mjpg/video.cgi?fps=5

Figure 10 : HTTP request to modify the frame per second

15

<parameter>=<value> Values Description

resolution=<string> 1280x1024, 1280x960, 1280x720, 768x576, 4CIF, 704x576, 704x480, VGA, 640x480, 640x360, 2CIFEXP, 2CIF, 704x288, 704x240, 480x360, CIF, 384x288, 352x288, 352x240, 320x240, 240x180, QCIF, 192x144, 176x144, 176x120, 160x120 1

Specify the resolution of the returned image.

camera=<string> 1, 2, 3, 4 or quad 1 Applies only to video servers with more than one video input. Selects the source camera.

compression=<int> 0 - 100 1 Adjusts the compression level of the image. Higher values correspond to higher compression, i.e. lower quality and smaller image size.

colorlevel=<int>2 0 - 100 1 Sets level of color or grey-scale.

0 = grey-scale, 100 = full color.

color=<int> 0, 1 Enables/disables color.

0 = black and white, 1 = color.

clock=<int> 0, 1 Shows/hides the time stamp.

0 = hide, 1 = show.

date=<int> 0, 1 Shows/hides the date.

0 = hide, 1 = show.

rotation=<int> 0, 90, 180, 270 1 Rotates the image clockwise.

textpos=<string> top, bottom The position of the string shown in the image.

overlayimage=<int>2 0, 1 Enable/disable overlay image.

overlaypos=<int>x<int>2 <xoffset>1x<yoffset>1 Set the position of the overlay image.

squarepixel=<int>2 0, 1 Enable/disable square pixel correction. Applies only to video servers.

Table 4 : HTTP parameters for dynamic management

16

5.3.2 TrainscopeViewer Management diagram

Figure 11 : TrainscopeViewer Management diagram

4 screens mode: Camera 1, 2, 3 and 4 Middle resolution Middle compression

Full screen mode: Camera 1 High resolution Low compression

HTTP requests: Stop video flow 2, 3 and 4 Camera1 resolution = high Camera1 compression = low

HTTP requests: Enable video flow 2, 3 and 4 Camera1,2,3,4 resolution = middle Camera1,2,3,4 compression = middle

16 screens mode: Camera 1, 2, 3 … 16 Low resolution Middle compression

HTTP requests: Enable video flow 5…16 Camera1…16 = resolution = low Camera1…16 compression = middle

HTTP requests: Disable video flow 5…16 Camera1,2,3,4 resolution = middle Camera1,2,3,4 compression = middle

HTTP requests: Stop video flow 2, 3 … 16 Camera1 resolution = high Camera1 = compression = low

HTTP requests: Enable video flow 5…16 Camera1…16 = resolution = low Camera1…16 compression = middle

17

6 Multi-point Accessing 6.1 Network design The TRAINET network is divided in two parts: land and onboard network. The system is illustrated in Figure 12 and Figure 13. In this design, the vehicle communicates to the “Hot-Spot” network on the land through the onboard network. The hot-spots transfer in real time the data to a router. This one regulates the data flow to the data server. In addition, dynamic bandwidth management software can automatically monitor the video stream.

Figure 12 : Onboard network diagram for TRAINET

Figure 13 : Land network diagram for TRAINET

Data server Computer Ex IP: 10.0.0.10

IP Router Ex IP: 10.0.0.3

Access Point Ex IP: 10.0.0.1

Power-over-Ethernet

Access Point N Ex IP: 10.0.0.N

Power-over-Ethernet

Access Point Ex IP: 10.0.0.2

Power-over-Ethernet

Wireless Bridge Ex IP: 10.0.0.13

5VDC

CAMERA 1 Ex IP: 10.0.0.11

CAMERA N Ex IP: 10.0.0.N

SWITCH 7.5VDC

CAMERA 3 Ex IP: 10.0.0.14

CAMERA 2 Ex IP: 10.0.0.12

18

6.2 Signal Handover

The bandwidth available depends on the network technology and the protocols used to manage the traffic. However, to understand entirely the bridge management our approach can not be limited by the consideration of the only link between one wireless bridge and one access point. The cooperation between the access points generates better results in terms of range and throughput. The drawbacks of this kind of approach are related to the protocols used while switching between access points. The previous generation of access points, they were waiting for the complete loss of the signal before trying to connect to another access point. The data flow was thereby irregular and could not be applied to cutting edge wireless applications.

Figure 14 : Network design for access point collaboration The last generation of access points allows for defining the data rate parameters. So the threshold data rate can be set to a value T. The access points check all the connections with the bridge (multipoint) superior to the T value. Then it chooses the best link among all the links available. The access points act in a “best effort” protocol to enhance the throughput. The data rate settings are used to choose the data rates that the access points use for data transmission. The rates are expressed in megabits per second. The access point always attempts to transmit at the highest data rate set to Basic, also called Require on the browser-based interface. If there are obstacles or interference, the access point steps down to the highest rate that allows data transmission. Data rate (1, 2, 5.5, and 11 megabits per second) can be set to one of three states:

1. Basic (this is the default state for all data rates)—Allows transmission at this rate for all packets, both unicast and multicast. At least one of the access point's data rates must be set to Basic.

2. Enabled—The access point transmits only unicast packets at this rate; multicast packets are sent at one of the data rates set to Basic.

AP AP AP AP

Server

70-110 meters

70-110 meters

70-110 meters

19

3. Disabled—The access point does not transmit data at this rate

In this example, the data rate threshold is equal to 12 Mbt/s (the average bandwidth for 4 cameras is 8 Mbts/s). The configuration settings of the different throughput are described below.

Network Throughput

Disabled Basic

Enabled

1/2/5.5/6 Mbps

9 Mbps

11 Mbps

12 Mbps

18 Mbps

24 Mbps

36 Mbps

48 Mbps

54 Mbps Table 5 : Configuration of the access point to avoid handovers

The threshold must be chosen according to the average bandwidth used in the network. If the threshold is too high, the access point will be unable to find a connection with the bridge. To solve this issue, the access points have to be set closer to each other in order to have a link with the bridge. However, it decreases the coverage of the wireless network. The figure 15 shows that the connection between the access points switches later with a 11Mbps threshold.

20

0

5

10

15

20

25

30

0 10 20 30 40 50 60 70 80 90 100 110 120 130

Distance meters

Ban

dwid

th M

bps

Threshold 11 MbpsThreshold 15 MbpsPoly. (Access Point 1)Poly. (Access Point 2)

Figure 15 : Signal Handover for 11 and 15 Mbps Threshold

AP1location: 30 meters AP2 location: 110 meters

21

6.3 Multiple bridge access points for multiple trains In a multiple bridge configuration, several transit shuttles establish a communication with the access points on the land network. In a point-to-multipoint configuration, two or more non-root bridges associate to a root bridge. According to the wireless expert Cisco, up to 17 non-root bridges can associate to a root bridge, but the non-root bridges must share the available bandwidth. Therefore, the maximum bandwidth available from the bridge must be shared between all the access points. Thanks to the signal handover management, the access point establishes the best link as possible with the bridge (Fig. 16). However, the point to multipoint range is inferior as the point to point range (Fig. 17).

Figure 16 : Multiple bridge access points for multiple trains

The bridge can be connected on different channels to avoid interferences and increase the bandwidth. Two pairs of bridges can be set up to add redundancy or load balancing to the bridge link. The bridges must use non-adjacent, non-overlapping radio channels to prevent interference, and they must use Spanning Tree Protocol (STP) to prevent bridge loops. The point to multipoint protocol “Frame relay” is used for the access points (Cisco Aironet 1300 series).Frame Relay is a networking protocol that works at the bottom two levels of the OSI reference model: the physical and data link layers. It is an example of packet-switching technology, which enables end stations to dynamically share network resources. It provides global addressing, Virtual Circuit status messages and multicasting. Frame Relay devices fall into the following two general categories:

• Data terminal equipment (DTEs), which include terminals, personal computers, routers, and bridges

AP 1 AP 2 AP 3 AP n

Airport Shuttle 1

Airport Shuttle 2

Airport Shuttle n

22

• Data circuit-terminating equipment (DCEs), which transmit the data through the network and are often carrier-owned devices (although, increasingly, enterprises are buying their own DCEs and implementing them in their networks)

Frame Relay networks transfer data using one of the following two connection types:

• Switched virtual circuits (SVCs), which are temporary connections that are created for each data transfer and then are terminated when the data transfer is complete (not a widely used connection)

• Permanent virtual circuits (PVCs), which are permanent connections

The DLCI is a value assigned to each virtual circuit and DTE device connection point in the Frame Relay WAN. Two different connections can be assigned the same value within the same Frame Relay WAN—one on each side of the virtual connection.

23

7 Distance issues For long network over 300 meters, the handover management is an important factor. However, the antennas are also crucial components to provide the best range and throughput. Two different high power antennas can be used: omnidirectionnal and directional antennas. In this investigation, only omnidirectionnal antennas for 802.11g are used.

0

5

10

15

20

25

30

0 10 20 30 40 50 60 70 80Distance from access point in meters

Net

wor

k tra

ffic

in M

egab

it pe

r sec

ond

Antenna 3dbiAntenna 10dbi

Figure 17 : Network traffic vs. distance

The graphic above (Fig.16) compares the throughput versus the distance in an outdoor environment for two antennas. The first antenna (3dbi) is a standard antenna and the throughput decreases quickly after 20 meters. The 10dbi omnidirectionnal antenna has a better range and the throughput decreases only after 50 meters. The maximum throughput observed is 26 Mbps (Fig 17) at a distance of 10 meters from the access point.

Figure 18 : Maximum bandwidth for outdoor conditions 26 Mbps

24

90 meters 45 meters

22 Mbps 10 Mbps

The throughput is directly linked to the linking quality of the bridge. It depends on the distance, the weather and the ‘line of sight’. The figure 18 represents the linking quality in an outdoor environment with the “line of sight”.

Figure 19 : Linking Quality vs. Distance from access points

There is a difference between the values claimed by the manufacturer (Table 6) and the values observed during our tests. The antenna range observed is represented Figure 19. The area within 45 meters from the access point has a throughput over 22 Mbps. Only this area of interest for our application.

Range 802.11g: Outdoor

110 ft (34m) @ 54 Mbps 200 ft (61 m)@ 48 Mbps 225 ft (69 m) @ 36 Mbps 325 ft (99 m) @ 24 Mbps

400 ft (122 m) @ 18 Mbps 475 ft (145 m) @ 12 Mbps 490 ft (149 m) @ 11 Mbps 550 ft (168 m) @ 9 Mbps 650 ft (198 m) @ 6 Mbps

660 ft (201 m) @ 5.5 Mbps 690 ft (210 m) @ 2 Mbps 700 ft (213 m) @ 1 Mbps

Table 6 : Network range for aironet 1200 Access Point (Cisco)

Figure 20 : Coverage of the omni directional antenna 10 dbi

25

As with all radio systems, interference is always a problem. If we are listening to an AM radio and we hear static, this is interference. The same thing applies to WiFi systems, however not to such a large degree. Things that cause interference with WiFi systems are Microwave ovens, certain lighting systems, other 802.11 access points or systems, microwave transmitters, even high speed processors for computers can cause interference for 802.11 systems. All these problems must be isolated before we can expect any significant range out of our system.

The use of directional antenna depends on the network’s design. It can increase significantly the range and the throughput for specific designs and improve the security of the network. At the beginning and at the end of the track, directional antenna should be set up to avoid transmitting the signal outside of the track where it could be intercepted.

26

8 Video Quality

Image quality is clearly one of the most important features of any camera, if not the most important. This is particularly so in surveillance and monitoring applications, where lives and property may be at stake. By developing image processing chips and sophisticated algorithms tailored for network camera applications, image quality has been improved to a degree never before seen at lower cost levels. As digital technology becomes commonplace and replaces analog solutions, there will be further advances in areas such as high resolution and advanced video compression, but success will ultimately depend on how well the initial information is captured and handled.

Superior image quality enables the user to:

• more closely follow details and changes in images, making for better and faster decisions concerning the safety of people and property

• With greater accuracy use automated analysis and alarm tools, such as face recognition, with fewer false positives.

Unlike traditional analog cameras, digital network cameras are equipped with the processing power not only to capture and present images, but also to digitally manage and compress them for network transport. Image quality can vary considerably and is dependent on the choice of optics and image sensor, the available processing power and the level of sophistication of the algorithms in the processing chip.

Image sensor

There are two types: CCD (Charged Coupled Device) and CMOS (Complementory metal oxide semiconductor) CCD sensors are produced using a technology developed specifically for the camera industry, while CMOS sensors are produced by the same technology used for the chips used in computers. Today's high quality cameras mostly use CCD sensors. Although recent advances in CMOS sensors are closing the gap, they are still not suitable for cameras where the highest possible image quality is required.

Image resolution

Higher resolution means more detail, and as cameras now deploy megapixel sensors that make it possible to capture even more detail, analog CCTV cameras--which are bound to resolutions used in TV standards--are being surpassed.

27

Vehicle 1 – Camera 1 Vehicle 1 – Camera 2

Vehicle 2 – Camera 3 Vehicle 2 – Camera 4 Figure 21 : Image quality with backlight compensation

Backlight compensation While a camera's automatic exposure control tries to get the lightness of an image to appear as the human eye would see a scene, it can be easily fooled. Think of the case where a person walks into a fairly dark room with a flashlight in her hand and directs this flashlight to the camera. Although the light source is quite small, it makes the camera believe the scene has become brighter and the camera's exposure control automatically adjusts to it, resulting in a darker image. To avoid this, a mechanism called backlight compensation is introduced. It strives to ignore small areas of high illumination, just as if they were not present at all. With backlight compensation, the image from the example above would have the same exposure regardless of whether the flashlight was present or not. The resulting image enables the person to be visually seen and identified. Without backlight compensation, the image would be too dark, and identification would be impossible.

Ability to correctly capture moving objects In addition to good light sensitivity, another key feature to look for is progressive scan. That the camera has progressive scan means that images do not suffer from the "saw" effect that hampers interlaced video technologies The interlace mode is used in TVs and traditional analog CCTV cameras in order to enhance the image frequency in moving images. The "saw" effect becomes apparent when the picture is frozen. Soon, network cameras and IP-Surveillance technology will deliver superior image quality by means of mega-pixel resolution. Analog cameras are limited by the 0.4 Megapixel resolution of NTSC/PAL standards.

28

9 Scalability 9.1 Storage

When the basis for the IP networking architecture was developed in the 1960s and '70s, the ability to provide redundancy was the top requirement. In the same way today, transmission links, application servers, storage and switches can all have parallel layers of services and alternative routes of communications.

Storage can be consolidated to secure off-site locations, and servers can use redundant power supplies, hot-swap RAID disks, error-correcting memory and dual network cards. This is all up to the network designer, and although a small network will not deploy all of the possible safety measures, choosing high-quality IT components in the network is in any case likely to be a more reliable solution than CCTV with VCRs or black box DVRs. And don't forget, by using standard server and network equipment, replacing faulty hardware takes much less time and is less costly than with proprietary DVR solutions.

Scalability steps: A DVR system is usually supplied with 4, 9 or 16 camera inputs, therefore becomes scalable in steps of 4, 9 or 16. If a system includes 15 cameras, this is not an issue, but it becomes a problem if 17 cameras are needed. Adding one single camera would generate the need for an additional DVR. Network video systems are far more flexible, and can be scaled in steps of one camera at a time.

Number of cameras per recorder: In a network video system, a PC server records and manages the video. The PC server can be selected according to the performance needed. Performance is often specified as number of frames per second, total for the system. If 30 fps is needed for each camera, one server may only record 25 cameras. If 2 fps is sufficient, 300 cameras can be managed by one server. This means that the performance of the system is used efficiently and can be optimized.

Size of system: For larger installations, a network video system is easy to scale. When higher recording frame rates or longer recoding times are needed, more processing and/or memory capacity can be added to the PC server managing the video. Even more simply, another PC server can be added, located either at a central location, or at remote locations.

9.2 Network Hardware 9.2.1 Digital Cameras scalability The system was tested with up to 5 digital cameras, but the number of cameras in a vehicle has no limit (contrary to an analog system). It only depends on the number of ports on the switch (5-80). The limiting factor is the bridge connection while some peaks up 26 Mbps were observed, only the average bandwidth of 22 Mbps will be considered in this study.

29

0

5

10

15

20

25

0 11 22 33 44 55 66 77 88 99 110

121

132

143

154

165

176

187

198

Number of cameras

Band

wid

th M

bps

240x180320x240640x480

Figure 22 : Bandwidth vs. Number of camera per vehicle at 1 fps for different resolution

A vehicle is equipped with several cameras (compression 30%, 1 fps) and tested with three different resolutions (240x180, 320x240, 640x480). With one bridge, the bandwidth can be shared up to 4 cameras in 640x480, up to 8 in 320x240 and up to 14 in 240x180.

0

5

10

15

20

25

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Number of cameras

Ban

dwid

th M

bps

240x180

320x240

640x480

Figure 23 : Bandwidth vs. Number of camera per vehicle at 30 fps for different resolution

A vehicle is equipped with several cameras (compression 30%, 30 fps) and tested with three different resolutions (240x180, 320x240, 640x480). With one bridge, the bandwidth can be shared up to 4 cameras in 640x480, up to 8 in 320x240 and up to 14 in 240x180.

4 cameras

4 cameras

30

0

5

10

15

20

25

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15

Number of cameras

Ban

dwid

th M

bps 240x180

320x240

640x480

Figure 24 : Bandwidth vs. Number of camera per vehicle at 15 fps for different resolution

A vehicle is equipped with several cameras (compression 30%, 15 fps) and tested with three different resolutions (240x180, 320x240, 640x480). With one bridge, the bandwidth can be shared up to 8 cameras for a resolution of 640x480.

0

5

10

15

20

25

0 18 36 54 72 90 108

126

144

162

180

198

Number of cameras

Ban

dwid

th M

bps compression

50%compression30%compression10%

Figure 25 : Bandwidth vs. Number of camera per vehicle at 1 fps for different compressions

A vehicle is equipped with several cameras (resolution 320x240, 30 fps) and tested with three different compressions (10%, 30%, 50%). With one bridge, the bandwidth can be shared up to 6 cameras in 640x480, up to 9 in 320x240 and up to 11 in 240x180.

4 cameras

4 cameras

31

0

5

10

15

20

25

0 1 2 3 4 5 6 7 8 9 10 11

Number of cameras

Ban

dwid

th M

bps compression

50%compression30%compression10%

Figure 26 : Bandwidth vs. Number of camera per vehicle at 30 fps for different compressions

A vehicle is equipped with several cameras (resolution 320x240, 30 fps) and tested with three different compressions (10%, 30%, 50%). With one bridge, the bandwidth can be shared up to 6 cameras in 640x480, up to 9 in 320x240 and up to 11 in 240x180.

0

5

10

15

20

25

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14

Number of cameras

Ban

dwid

th M

bps compression

50%compression30%compression10%

Figure 27 : Bandwidth vs. Number of camera per vehicle at 15 fps for different compressions

A vehicle is equipped with several cameras (resolution 320x240, 15 fps) and tested with three different compressions (10%, 30%, 50%). With one bridge, the bandwidth can be shared up to 12 cameras in 640x480.

4 cameras

4 cameras

32

9.2.2 Increasing the length of the track

Access Point* Switch Ethernet cable (90 meters)

Optical Switch (number of ports)

<280 meters 3 1 3 No optical <540 meters 6 2 6 7 <810 meters 9 3 9 10 <1080 meters 12 4 12 13 <1350 meters 15 5 15 16 <1620 meters

= 1mile 18 6 18 19

Table 7 : Network configuration for several network range* *The number of access points is available for an omnidirectional antenna 3dbi. The use of different antennas (directional, 10 dbi) would reduce this number. The table 7 describes the scalability of the system for different lengths of the track. For a length under 280 meters, Ethernet cable can be used instead of optical fiber. For longer track, the optical switch can interconnect the different access points. Moreover, the optical fiber has a bandwidth available equal to several Giga bits per seconds. 9.2.3 Increasing the number of vehicles

Figure 28 : Bandwidth vs. Number of vehicle at 30 fps 320x240 30%

Theoretically, each vehicle is equipped with four or six cameras (resolution 320x240, 15 fps, compression 30%). In a point to multipoint mode, the vehicles share the available bandwidth on the same channel. With vehicles equipped with six cameras, the system can manage two vehicles and up to three vehicles equipped with four cameras. This is the maximum number connection per access point. More vehicles can be added but they do not have to connect the same access point at the same time.

0

10

20

30

40

50

60

70

1 2 3 4 5 6

Number of vehicle

Band

wid

th M

bps

4 cameras6 cameras

22 Mbps

33

0

1

2

3N

umbe

r of

brid

ges

1 10 20 30 40 50 60 70 80 90 100

4 cameras6 cameras

Number of vehicles4 cameras

6 cameras

Figure 29 : Number of bridges vs. number of vehicles (4 cameras or 6 cameras) at 1 fps

0

1

2

3

Num

ber

of

brid

ges

1 2 3 4 5 6

4 cameras6 cameras

Number of vehicles 4 cameras

6 cameras

Figure 30 : Number of bridges vs. number of vehicles (4 cameras or 6 cameras) at 30 fps

A way to tackle this problem is to add several bridges. Two pairs of bridges can be set up to add redundancy or load balancing to the bridge link. The bridges must use non-adjacent, non-overlapping radio channels to prevent interference, and they must use Spanning Tree Protocol (STP) to prevent bridge loops. The figure 26 shows the theoretical results for vehicle equipped with four or six cameras (resolution 320x240, 30 fps, compression 30%). The figure 27shows the theoretical results for vehicle equipped with four or six cameras (resolution 320x240, 15 fps and compression 30%).

34

0

1

2

3

Num

ber o

f br

idge

s

1 2 3 4 5 6 7 8

4 cameras

6 cameras

Number of vehicles4 cameras6 cameras

Figure 31 : Number of bridges vs. number of vehicles (4 cameras or 6 cameras) at 15 fps

9.3 Pittsburgh Technology Center test for throughput Several tests have been realized in the Pittsburgh Technology Center (PTC) for measuring the throughput. The goal was to reproduce the results obtained in the indoor laboratory. The parameters such as weather, line of sight and vehicle speed were unknown factors for the network throughput. The network is configured as described Figure 32 and Figure 33.

Figure 32 : Onboard network diagram for PTC TEST

Wireless Bridge IP: 10.0.0.13

CAMERA 1 IP: 10.0.0.11

CAMERA 4 IP: 10.0.0.15

SWITCH

CAMERA 3 IP: 10.0.0.14

CAMERA 2 IP: 10.0.0.12

35

Figure 33 : Land network diagram for PTC TEST

The test road is 300 meters long (Fig. 36 yellow line) and its configuration is close to the design of the test track in Bombardier Transportation. Two access points are set up along the track (Fig. 36 white points) to cover the all distance. The access points have a default configuration and all the different network throughput are enabled.

Data server Computer IP: 10.0.0.13

IP Router Ex IP: 10.0.0.3

Access Point Ex IP: 10.0.0.1

Power-over-Ethernet

Access Point Ex IP: 10.0.0.2

Power-over-Ethernet

36

Figure 34 : Camera set up 1

Figure 35 : Camera set up 2

Table 8 : Configuration of the access points (Access Point 1 & 2)

Network Throughput

DiCamera set uosabled

Basic

Enabled

1/2/5.5/6/9 Mbps

11 Mbps

12 Mbps

18 Mbps

24 Mbps

36 Mbps

48 Mbps

54 Mbps

37

Figure 36 : Satellite view Pittsburgh Technology Center test Yellow line: test road (≈300 meters)

White Points: Access points

The resulting coverage is about 220 meters. However, two zones must be differentiated: the middle zone and the border zone. The handover is smooth between the access points and the throughput is about 24 Mbps (Table 9). The configuration is really efficient in the middle zone and the results match the criteria of the project in terms of resolution, number of cameras, distance and throughput. The throughput used for the video is under the half of the bandwidth available (43%). The bandwidth available is shifting due to the handover, this issue is solved in the next chapter. The connection in the border zone is poor and some disconnections happen at the very end of the track. The frame rate is slow (10 fps) but the quality of the picture is still good due to the MJPEG compression. Each frame has the quality of a JPEG image. Middle Zone

(between access points)

Border Zone (between access point and end of the track)

Frame per second 30 10 Distance from access

point 40 50

Resolution 320x480 320x480 Compression 10% 10%

Bandwidth available 24 Mbps 14 Mbps Bandwidth used 10 Mbps (43%) 5 Mbps (22%)

Delay No Yes Disconnection No Yes at the end

Table 9 : Network Measures for the Pittsburgh Technology Center

38

Number of cameras 4 Number of access points 2

Number of bridge 1 Bandwidth 14 -24 Mbps Resolution 320x480 Antenna 10dbi

Compression 10% Distance covered with network connection 250 meters Distance covered with throughput >22 Mps 190 meters

Vehicle speed 40-50 mph Table 10 : Results for the Pittsburgh Technology Center

9.4 Pittsburgh Technology Center Test for Handover The onboard network remains the same as for the test for the throughput. An access point is added to the land network in order to test the reliability of the handover (Fig. 37). Three access points are set up along the track at the distance 60, 140 and 220 meters (Fig. 38). The configuration of the access points (Fig. 38) were changed to maximize the throughput when switching between access points. The threshold T is set gradually to 11, 12, 18 Mbps.

Figure 37: Land network diagram for PTC TEST

Data server Computer IP: 10.0.0.13

IP Router Ex IP: 10.0.0.3

Access Point Ex IP: 10.0.0.1

Power-over-Ethernet

Access Point Ex IP: 10.0.0.4

Power-over-Ethernet

Access Point Ex IP: 10.0.0.2

Power-over-Ethernet

39

Network Throughput Disabled Basic Enabled

1/2/5.5/6/9 Mbps

11 Mbps

12 Mbps

18 Mbps

24 Mbps

36 Mbps

48 Mbps

54 Mbps

Table 11 : Configuration of the access points for a threshold equal to 11Mbps

Figure 38 : Satellite view Pittsburgh Technology Center test

Yellow line: test road (≈300 meters) White Points: Access points

The throughput is measured along the track for the different thresholds (Fig. 39) and the disconnection are indicated with bandwidth available equal to zero. Therefore, a threshold at 11 Mbps has a better coverage (320 meters but the average bandwidth is 17-20 Mbps). For the 12 Mbps threshold, the average bandwidth is 20-22 Mbps for coverage of 280 meters whereas for the 18 Mbps, the coverage is only 200 meters but with an average bandwidth of 21-23 Mbps. The investigators observed a disconnection at the distance 180 meters of the track. This drawback is due to the threshold which is too high. Neither of the access point can find a link with the bridge with a 18 Mbps threshold. That is the reason why a threshold of 12 Mbps will be used in the tests. It is the best compromise between coverage, throughput and handover.

40

Handover management for different thresholds

0

5

10

15

20

25

0 20 40 60 80 100

120

140

160

180

200

220

240

260

280

300

Test track distance (meters)

Ban

dwid

th (M

bps)

11 Mbps12 Mbps18 Mbps

Figure 39 : Bandwidth vs. distance (thresholds 11 Mbps, 12 Mbps and 18 Mbps)

Number of cameras 4 Number of access points 3

Number of bridge 1 Bandwidth 20-22 Mbps

Threshold for handover 12 Mbps Resolution 320x480

Compression 10% Distance covered with throughput >22 Mps 280meters

Vehicle speed 40-50 mph Table 12 : Results for the Pittsburgh Technology Center – Handover test

41

9.5 Pittsburgh Technology Center test for multiple bridges Three access points are set up along the track at the distance 60, 140 and 220 meters (Fig.40 and Fig.41). The access points are set up in point to multipoint mode. The land network remains the same as the test for the handover is concerned. There are two mobile onboard networks. Each one includes two IP cameras fastened to the car frames. Two scenarios are tested: 1.The cars follow each other in order to share the same connection (same access point). 2.The cars are coming from two different directions and intercept in the middle of the test

track.

Figure 40 : Onboard network for vehicle 1 Figure 41 : Onboard network for vehicle 2

9.5.1 Scenario 1: Cars following each other In this scenario, each access point has to handle two bridges connection. The video quality is the same as the point to multipoint connection (Image Quality 320x480, Compression 10%). The vehicles follow each other at a speed of 20-25 mph and use the road track of the Pittsburgh Technology Center (Fig. 42).

Wireless Bridge IP: 10.0.0.19

SWITCH

CAMERA 4 IP: 10.0.0.14

CAMERA 3 IP: 10.0.0.13

Wireless Bridge IP: 10.0.0.13

SWITCH

CAMERA 2 IP: 10.0.0.12

CAMERA 1 IP: 10.0.0.11

42

Vehicle 1 – Camera 1 Vehicle 1 – Camera 2

Vehicle 2 – Camera 3 Vehicle 2 – Camera 4

Figure 42 : Camera screenshots for multiple bridges scenario 1

Figure 43 : Multiple bridges with two vehicle following each other

The frame rate is 30 frames per second between two access points but decreases quickly when the two bridges are connected to the same access point (Fig. 43). The throughput received by the network is 10 Mbps and is the same as in a point to point mode. To maintain, a frame rate equal to 30 frames per second, the vehicle must always be between two access points. When the two bridges are connected to the same access point, that

AP 1 AP 2 AP 3

Vehicle 2

Vehicle 1

Vehicle 1

Vehicle 2

30 Frames per second Image Quality 320x480

Compression 10%

10 Frames per second Image Quality 320x480

Compression 10%

43

point must handle all the communication and it introduces a delay in the video flow processing.

Number of cameras 2x2 Number of access points 3

Number of bridge 2 Antenna 10 dbi

Bandwidth 20-22 Mbps Threshold for handover 12 Mbps

Resolution 320x480 Compression 10%

Distance covered with throughput >22 Mps 230meters Vehicle speed 40-50 mph

Frame per seconds 30 but only 10 at the end of the track

Table 13 : Results for the Pittsburgh Technology Center – multiple bridges test1

9.5.2 Scenario 2: Cars intercepting The video quality is the same as the point to multipoint connection (Image Quality 320x480, Compression 10%). The vehicles intercept each other at a speed of 30-40 mph and use the road track of the Pittsburgh Technology Center (Fig. 45). The cars start from the two extremity of the track and drive in an opposite direction to intercept in the middle of the track.

Vehicle 1 – Camera 1 Vehicle 1 – Camera 2

Vehicle 2 – Camera 3 Vehicle 2 – Camera 4

Figure 44 : Camera screenshots for multiple bridges scenario 1

44

Figure 45 : Multiple bridges with two vehicles intercepting

This test is close to the real application with two airport shuttles making round-trips between the airport and the planes. At the time T=0s, the connection is point to multipoint and the bandwidth available for each connection is 22 Mbps. In this test, 30 frames per seconds are observed from T=0s to T=25s. When the vehicles intercept in the middle of the track (T=25s to T=30s), a brief slowdown (<2s) is observed. In the last test, 30 frames per seconds are observed in the point to multipoint mode. This value confirms the results of the previous test with two vehicles following each other between two access points.

Number of cameras 2x2 Number of access points 3

Number of bridge 2 Antenna 10 dbi

Bandwidth 20-22 Mbps Threshold for handover 12 Mbps

Resolution 320x480 Compression 10%

Distance covered with throughput >22 Mps 280meters Vehicle speed 30-40 mph

Frame per seconds 30 but only 10 at the end of the track

Table 14 : Results for the Pittsburgh Technology Center – multiple bridges test2

AP 1 AP 2 AP 3

Vehicle 1 Vehicle 2

30 Frames per second Image Quality 320x480

Compression 10%

Time T=0s

Time T=30s

AP 1 AP 2 AP 3

Vehicle 1

Vehicle 2

30 Frames per second Image Quality 320x480

Compression 10%

45

9.6 Bombardier Transportation Test 9.6.1 System Configuration The final test was realized on the Bombardier Transportation’s test track in Pittsburgh. The design of the track is presented below (Fig. 46). The track is 300 meters long. For this distance, three access points are set up along the track (AP1:55 meters, AP2: 145 meters, AP3: 235 meters).

Figure 46 : Bombardier Transportation test track satellite view

Yellow line: test road (≈300 meters) White Points: Access points

Figure 47 : Shuttle vehicle on the test track

46

9.6.2 Network settings The network is configured to switch when the connection bandwidth is under 12Mbit/s. The cameras located in the moving vehicle are managed dynamically by the software “TrainscopeViewer”.

Figure 48 : Onboard network diagram for TRAINET

Figure 49 : Land network diagram for TRAINET

9.6.3 Field results With the help of a camera recorder, the video flow is recorded. The flow is smooth and 30 frames per seconds are observed. However, some packets are lost during the wireless transfer. It results some slowdown while transferring the video. This drawback can be avoided while fastening the access points to a post. It will avoid most of the reflection with the ground. The image quality is good and we can clearly identify building and people moving inside the vehicle.

Data server Computer IP: 10.0.0.10

IP Router IP: 10.0.0.3

Access Point IP: 10.0.0.1

Access Point 3 IP: 10.0.0.4

Access Point IP: 10.0.0.2

Wireless Bridge Ex IP: 10.0.0.13

CAMERA 1 IP: 10.0.0.11

CAMERA 4 IP: 10.0.0.15

SWITCH

CAMERA 3 IP: 10.0.0.14

CAMERA 2 IP: 10.0.0.12

47

Figure 50 : Image recorded through the wireless network The frames are recorded through the wireless network with a resolution of 320x240 pixels and a compression of 50%. The format for the video is MJPEG, therefore the still images have the quality of JPEG. Some recorders can be added onboard to record the video flow for saving the video in high definition. Due to the dynamic bandwidth management only 10Mbps are used the worst case. The average bandwidth used is 8 Mbt/s and is far below the capacity of the wireless link (20-22Mbts). This low bandwidth is linked to the low complexity of the scene (see the last appendix of the published paper). Most of the cameras were aimed outside and half of the pictures represented the sky. The movement of the people in the vehicle will increase the complexity of the scene and by consequence the bandwidth used (Annexe 1D-1E).

Figure 51 : Network traffic for Bombardier Test track (average 8 Mbps) (100% 100Mbps)

48

Number of cameras 4 Number of access points 3

Number of bridge 1 Antenna 10 dbi

Bandwidth 20-22 Mbps Threshold for handover 11 Mbps

Resolution 320x480 Compression 10%

Distance covered with throughput >22 Mps

280meters

Vehicle speed 30-40 mph Frame per seconds 20-30 fps

Table 15 : Results for Bombardier Transportation Test

49

Fig. 45: Bombardier test track design

Start

Stop

Access Point 1 Access Point 2 Access Point 3

280 meters

50

Glossary of acronyms

16-QAM — 16-Phase Quadrature Amplitude Modulation 64-QAM — 64-Phase Quadrature Amplitude Modulation ADC — Analog-to-Digital Converter AES — Advanced Encryption Standard AHB — Advanced High-performance Bus APB — Advanced Peripheral Bus ARM-CPU — A Central Processing Unit based on Intellectual Property Licensed from ARM ATM — Asynchronous Transfer Mode BPSK — Binary Phase-Shift Keying BWA — Broadband Wireless Access CPE — Customer Premises Equipment CRC — Cyclic Redundancy Code DAC — Digital-to-Analog Converter DES — Data Encryption Standard DMA — Direct Memory Access DSL — Digital Subscriber Line E1 — a dedicated phone connection supporting data rates of 2 Mb/s ECC — Error Correction Code FDD — Frequency Division Duplexing FFT — Fast Fourier Transform IEEE — Institute of Electrical and Electronics Engineers IP — Internet Protocol LOS — Line-of-Sight MAC — Media Access Control MMDS — Multichannel Multipoint Distribution Service NLOS — Non-Line-of-Sight OFDM — Orthogonal Frequency Division Multiplexing PDA — Personal Digital Assistant PHY — Physical Layer PMP — Point-to-Multipoint QPSK — Quadrature Phase-Shift Keying RF — Radio Frequency RISC — Reduced Instruction Set Computer RX — Receiver SoC — System on Chip T1 — a dedicated phone connection supporting data rates of 1.544 Mb/s TDD — Time Division Duplexing TX — Transmitter

51

Hardware list SUMMARY

Component Task Power Vendor App. price Datasheet

Linksys WET54Gv2

Wireless bridge

10W 5V-DC / 2A Linksys $110 US 1-A

Netgear 5 Ports Switch

7.5W 7.5V-DC /

1A Netgear $50 US 1-B

Axis 210 Digital Camera 10W Axis $450 US 1-C

Table.1. Hardware list for the TRAINET’s onboard network

Linksys WET54Gv2 Netgear 5 Ports Axis 210

Component Task Power Vendor App. price Datasheet

IBM Thinkpad Data server Computer

16V-DC / 4.5A IBM $1410 US -

Cisco Catalyst 3500 series XL Router

7.5W 7.5V-DC /

1A Cisco $100 US 1-D

Cisco Aironet 1200 Series

Wireless Access Points

48V-DC / 380mA Cisco $450 US 1-E

Ethernet Cable 90 meters

Link components - Pittsburgh

Wires $50 US -

Table.1. Hardware list for the TRAINET’s onboard network

52

Cisco Aironet 1200 Series Cisco Catalyst 3500 series

XL

Ethernet Cable

53

Annexe 1A - Linksys WET54Gv2 Datasheet

Features

Product Description - Linksys Wireless-G Ethernet Bridge WET54G - wireless access point

Device Type - Wireless access point

Enclosure Type – External

Dimensions (WxDxH) - 5 in x 4.2 in x 1.1 in

Weight - 0.4 lbs

Data Link Protocol - Ethernet, Fast Ethernet, IEEE 802.11b, IEEE 802.11g

Network / Transport Protocol - TCP/IP

Features - MDI/MDI-X switch

OS Required - Microsoft Windows 98 Second Edition / Windows ME, Microsoft Windows 2000 / XP Technical Specs

Device Type - Wireless access point

Width - 5 in

Depth - 4.2 in

Height - 1.1 in

Weight - 0.4 lbs

Form Factor - External

Connectivity Technology - Wireless

Data Transfer Rate - 54 Mbps

Line Coding Format - DBPSK, DQPSK, CCK, OFDM

Data Link Protocol - Ethernet, Fast Ethernet, IEEE 802.11b, IEEE 802.11g

Network / Transport Protocol - TCP/IP

Frequency Band - 2.4 GHz

Status Indicators - Port status, power

Features - MDI/MDI-X switch

Encryption Algorithm - 128-bit WEP, 64-bit WEP

Compliant Standards - IEEE 802.3, IEEE 802.3U, IEEE 802.11b, IEEE 802.11g

Antenna - Detachable

Antenna Qty - 1

Interfaces - 1 x network - Ethernet 10Base-T/100Base-TX - RJ-45 1 x network - Radio-Ethernet

54

Cables Included - 1 x network cable

Compliant Standards - FCC Class B certified

Power Device - Power adapter - external

Software Included - Drivers & Utilities

OS Required - Microsoft Windows 98 Second Edition / Windows ME, Microsoft Windows 2000 / XP

Min Operating Temperature - 32 F

Max Operating Temperature - 104 F

Humidity Range Operating - 10 - 90%

55

Annexe 1B – NETGEAR SWITCH 5 PORTS

The FS100 series Fast Ethernet switches brings the 100 Mbps switching technology in a compact form factor to the small office marketplace at an aggressive price. These switches segment networks for improved performance, enabling the most demanding multimedia and imaging applications. Since each port is auto-speed-sensing, individual hubs or directly attached servers can easily be upgraded from 10 to 100 Mbps.

Each port can automatically negotiate the network speed and duplex mode with the remote end, taking the burden of configuration off the user. At the higher speed running in full duplex mode, each port supports up to 200 Mbps of information throughput. With it’s speed and it’s reliability, the FS100 series is the best choice for your networking needs.

Main Specs: Product Description: NETGEAR FS105 - switch - 5 ports Device Type: Switch Form Factor: External Dimensions (WxDxH): 6.2 in x 4.1 in x 1.1 in Weight: 1.3 lbs Ports Qty: 5 x Ethernet 10Base-T, Ethernet 100Base-TX Data Transfer Rate: 100 Mbps Data Link Protocol: Ethernet, Fast Ethernet Communication Mode: Half-duplex, full-duplex

Features: Full duplex capability, uplink, MDI/MDI-X switch

Compliant Standards: IEEE 802.3U-LAN, IEEE 802.3i-LAN, IEEE 802.3x

Manufacturer Warranty: 5 years warranty

56

Annexe 1C – AXIS 210 DIGITAL CAMERA

General • Motion JPEG or MPEG-4 based network camera with built-in web server • Superior quality CCD image sensor with progressive scan

Security • Multi user level password protection for restricted camera access • IP address filtering

System Requirements The following specification applies to the use of browser-based viewing. When using application software, please refer to the specifications provided with the software. • Supported operating systems: Windows XP, 2000, NT4.0, ME and 98, Linux and Mac OS X • Supported Web Browsers: Windows - Internet Explorer 5.x or later and Mozilla 1.4* or later Linux - Mozilla 1.4* or later Mac OS X - Mozilla 1.4* or later and Netscape 7.1* or later • Hardware: Meeting the specification for selected operating system and browser

Hardware & System • AXIS ETRAX 100LX 32bit RISC CPU • 16 MB RAM, 4 MB Flash • AXIS ARTPEC-2 video compression chip with 8 MB RAM • Linux 2.4 kernel • Watch Dog functionality

Connections • Network: 10Base-T/100Base-TX Ethernet networks (RJ-45) • I/O: 1 alarm input + 1 output (terminal block) • Power: 9 VDC / 9 W - external Power Supply, included • Alternative input voltage 7-20V DC, min 4W

Camera • Lens F2.0 4 mm glass lens, C/CS mount • Focus range 0.5 mm to infinity • Sony Super HAD progressive scan CCD 1/4” RGB • Color & Black/White • Frame rate: up to 30 frames per second in all resolutions • Video compression: Motion JPEG and MPEG-4 (Part 2, Advanced Simple Profile at level 5) • Supports various resolutions and video quality settings • Max resolution: 640 x 480 VGA in 30 frames per second • Illumination: 3-10 000 Lux • Rotation: 90, 180, 270 degrees • Image scaling The file size of a JPEG image depends on resolution, compression level and image content. Below is a table with average file size in KByte, derived from real life tests

Functions • Built-in Video Motion Detection • Scheduled and triggered event functionality with alarm notification via e-mail, TCP, HTTP and upload of images via e-mail, FTP & HTTP • Pre/post alarm buffer of up to 20 seconds of 320x240 resolution video at 4 frames per second • A mask, custom logo or overlay may be inserted in the video stream • Flash memory provide ability to upload embedded applications • Up to 20 simultaneous users • UPnP • AXIS Dynamic DNS service Firmware updates • Flash memory allows firmware updates over the network using HTTP or FTP over TCP/IP. Firmware upgrades are available from www.axis.com

57

Annexe 1D – Cisco Catalyst 3500 series XL

Description Specification Performance 10.8 Gbps switching fabric

8.8 million packets-per-second forwarding rate (Catalyst 3548 XL) 7.5 million packets-per-second forwarding rate (Catalyst 3508G XL)6.5 million packets-per-second forwarding rate (Catalyst 3524 XL) 4.8 million packets-per-second forwarding rate (Catalyst 3512 XL) (All forwarding rates for 64-byte packets) 5.4 Gbps max forwarding bandwidth 4 MB shared-memory architecture shared by all ports Packet forwarding rate for 64-byte packets: 14,880 PPS to 10 Mbps ports 148,800 PPS to 100BaseT ports 1,488,000 PPS to 1000BaseX ports 8192 Media Access Control (MAC) addresses 8 MB DRAM and 4 MB Flash memory onboard

Management IEEE 802.3x full duplex on 10BaseT, 100BaseTX, and 1000BaseX ports IEEE 802.1D Spanning-Tree Protocol IEEE 802.1Q VLAN IEEE 802.3z 1000BaseX specification 1000BaseX (GBIC) 1000BaseSX 1000BaseLX/LH IEEE 802.3u 100BaseTX specification IEEE 802.3 10BaseT specification

Y2K Y2K compliant Connectors and Cabling

10BaseT ports: RJ-45 connectors; two-pair category 3, 4, or 5 unshielded twisted-pair (UTP) cabling 100BaseTX ports: RJ-45 connectors; two-pair Category 5 UTP cabling 1000BaseX GBIC ports: SC fiber connectors, single mode or multimode fiber GigaStack GBIC ports: copper-based Cisco GigaStack cabling Management console port: RJ-45 connector, RS-232 serial cabling

Indicators Per-port status LEDs - link integrity, disabled, activity, speed, and full-duplex indications System status LEDs - system, RPS, and bandwidth utilization indications

Warranty Note All units include a lifetime return-to-factory warranty

58

Annexe 1E - Cisco Aironet 1200 Series

Item Specification

Part Number

• AIR-AP1231G-x-K9 • Regulatory Domains: (x = Regulatory Domain) • A = FCC • E = ETSI • I = Israel • J = TELEC (Japan)

Customers are responsible for verifying approval for use in their individual countries. To verify approval and to identify the regulatory domain that corresponds to a particular country please visit: http://www.cisco.com/go/aironet/compliance Not all regulatory domains have been approved. As they are approved, the part numbers will be available on the Global Price List.

Software Cisco IOS Software

Data Rates Supported 802.11g: 1, 2, 5.5, 6, 9, 11, 12, 18, 24, 36, 48, and 54 Mbps

Network Standard IEEE 802.11b and IEEE 802.11g

Uplink Autosensing 802.3 10/100BASE-T Ethernet

Radio Module Form Factor • 802.11a: CardBus (32-bit) • 802.11b or 802.11g: Mini-PCI (32-bit)

Frequency Band and Operating Channels

Americas (FCC) 2.412 to 2.462 GHz; 11 channels ETSI 2.412 to 2.472 GHz; 13 channels Israel 2.432 to 2.472 GHz; 9 channels

Japan (TELEC) 2.412 to 2.472 GHz; 13 channels Orthogonal Frequency Division Multiplexing (OFDM) 2.412 to 2.484 GHz; 14 channels Complementary Code Keying (CCK)

Nonoverlapping Channels 802.11g: 3

59

Wireless Modulation 802.11g: Direct sequence spread spectrum (DSSS); OFDM

Receive Sensitivity (Typical)

802.11g: 6 Mbps: -90 dBm 9 Mbps: -84 dBm 12 Mbps: -82 dBm 18 Mbps: -80 dBm 24 Mbps: -77 dBm 36 Mbps: -73 dBm 48 Mbps: -72 dBm 54 Mbps: -72 dBm

Available Transmit Power Settings (Maximum power setting will vary according to individual country regulations)

802.11g: CCK: 100 mW (20 dBm) 50 mW (17 dBm) 30 mW (15 dBm) 20 mW (13 dBm) 10 mW (10 dBm) 5 mW (7 dBm) 1 mW (0 dBm)

Range

802.11g: Outdoor 110 ft (34m) @ 54 Mbps 200 ft (61 m)@ 48 Mbps 225 ft (69 m) @ 36 Mbps 325 ft (99 m) @ 24 Mbps 400 ft (122 m) @ 18 Mbps 475 ft (145 m) @ 12 Mbps 490 ft (149 m) @ 11 Mbps 550 ft (168 m) @ 9 Mbps 650 ft (198 m) @ 6 Mbps 660 ft (201 m) @ 5.5 Mbps 690 ft (210 m) @ 2 Mbps 700 ft (213 m) @ 1 Mbps

802.11g: Indoor 90 ft (27 m) @ 54 Mbps 95 ft (29 m) @ 48 Mbps 100 ft (30 m) @ 36 Mbps 140 ft (43 m) @ 24 Mbps 180 ft (55 m) @ 18 Mbps 210 ft (64 m) @ 12 Mbps 220 ft (67 m) @ 11 Mbps 250 ft (76 m) @ 9 Mbps 300 ft (91 m) @ 6 Mbps 310 ft (94 m) @ 5.5 Mbps 350 ft (107 m) @ 2 Mbps 410 ft (125 m) @ 1 Mbps

Ranges and actual throughput vary based upon numerous environmental factors so individual performance may differ.

Compliance

Standards Safety

• UL 60950 • CAN/CSA C22.2 No. 60950 • IEC 60950 • UL 2043

Radio Approvals • FCC Part 15.247 • RSS-210 (Canada) • EN 300.328 • ARIB-STD 33 (Japan) • ARIB-STD 66 (Japan) • AS/NZS 4771 (Australia and New Zealand)

EMI and Susceptibility (Class B) • FCC Part 15.107 and 15.109 • ICES-003 (Canada) • VCCI (Japan) • EN 301.489-1 and -17 (Europe) • AS/NZS 3548

Security • 802.11i, WPA2, WPA • 802.1X

60

• AES, TKIP Other

• IEEE 802.11g • FCC Bulletin OET-65C • RSS-102

Antenna

2.4 GHz Radio: • Two RP-TNC connectors; 802.11g approved with:

–AIR-ANT1728, AIR-ANT1729, AIR-ANT2012, AIR-ANT2506, AIR-ANT3213, AIR-ANT3549, AIR-ANT4941, AIR-ANT5959, and AIR-ANT2410Y-R

Network Management

BootP, Secure Shell (SSH) Protocol, Secure HTTP (HTTPS), Trivial File Transfer Protocol (TFTP), FTP, Telnet, console port, Simple Network Management Protocol (SNMP) MIB I and MIB II, CiscoWorks Resource Manager Essentials (RME), CiscoWorks Software Image Manager (SWIM), CiscoWorks Campus Manager, CiscoWorks CiscoView, and CiscoWorks WLSE

LEDs Three indicators on the top panel report Ethernet activity and status, device operating status, and radio activity and status.

Housing Die-cast aluminum

Dimensions (H x W x D)

1.660 x 6.562 x 7.232 in. (4.22 x 16.67 x 18.37 cm); add 0.517 in. (1.31 cm) height for mounting bracket

Weight 1.725 lb (0.783 kg); add 0.4 lb (0.181 kg) for mounting bracket

Environmental -4 to 122°F (-20 to 50°C), 10 to 90 percent humidity (noncondensing)

Memory and Processor

IBM PowerPC405 (200 MHz) 16 MB RAM; 8 MB Flash memory

Input Power Requirements

90 to 240 VAC ±10 percent (power supply) 48 VDC ±10 percent

Power Draw 6W maximum

Warranty One year

Wi-Fi Certification

61

ANNEXE 2: Extended Antennas Options

Omnidirectional Antenna

An omnidirectional antenna works equally well in picking up signals from every direction. Omni's make excellent general purpose and mobile antenna's. The longer an omnidirectional antenna is, the better performance it will have.

Parabolic or Dish Antenna

A parabolic antenna will give you the greatest range for your signal. The trade-off is that they are more difficult to aim. A parabolic antenna is the obvious choice for a point-to-point fixed wireless installation.

Yagi Antenna

A Yagi antenna is a good choice for point-to-point transmission, or point-to-multipoint where the destination antenna are all close together or in a straight line.

Patch Antenna and Sector Antenna

The most common use of the patch and sector antenna designs is as the customer antenna in a point-to-point wireless broadband system.

62

Maximum Power Level (mW) Regulatory Domain Antenna Gain (dBi) CCK OFDM

5.2 (Omni) 100 30

9 (Patch) 100 30

10 (Yagi) 100 30

11 (Omni) — —

12 (Omni) 100 30

13 (Integrated patch) 100 30

13.5 (Yagi) 100 30

14 (Sector) 50 20

Americas (-A) (4 W EIRP maximum)

21 (Dish) 20 10

5.2 (Omni) 20 10

9 (Patch) 10 5

10 (Yagi) 10 5

11 (Omni) — —

12 (Omni) 5 1

13 (Integrated patch) 5 1

13.5 (Yagi) 5 1

14 (Sector) 1 1

EMEA (-E) (100 mW EIRP maximum)

21 (Dish) 11 —

5.2 (Omni) 10 10

9 (Patch) 10 10

10 (Yagi) 10 10

11 (Omni) 10 10

12 (Omni) 10 10

13 (Integrated patch) 10 10

13.5 (Yagi) 10 10

14 (Sector) 10 10

Japan (-J) (10 mW/MHz EIRP maximum)

21 (Dish) 10 10 Maximum Power Levels Per Antenna Gain for IEEE 802.11g

63

ANNEXE 3:Flow On Demand for Video Throughput Control

Guillaume Milcent and Yang Cai, Carnegie Mellon University,, ycai@cmu.edu, published on AmI-Life Workshop Proceedings, Spain, 2005 and to appear on LNAI 3864, 2006

Abstract. The bandwidth of the wireless network has been a bottleneck for live security video streaming. In this paper, we introduce the concept of the bandwidth of human visual processing. We intend to match the human information processing bandwidth with physical network bandwidth. An eye gaze-based interface is used to optimize the video flow by adjusting the video resolution and compression to reduce the bandwidth by up to 75%. With the addition of a face-based compression method the bandwidth is reduced by up to 88%.

1 Introduction

The digital video cameras have been increasingly used in surveillance systems. For wireless network the bandwidth is the limiting factor and displaying more than 10 video with high resolution, high frame rate and high quality is currently a big issue [10][11]. This problem is even more significant for a large, wired video surveillance system, which requires displaying hundreds of cameras. At the same time, Flow On Demand (FOD) for video throughput control is one way to tackle this problem. It calculates the flow required by the user according to the ways that the different screens are viewed. When a screen is viewed, the display becomes larger and the compression is reduced. This way more details of the picture currently displayed can be shown. Reducing the bandwidth has become a hot topic for research in recent years because of the over increasing size of digital networks especially in the medical and defense sectors [10][11][14]. Sending high resolution information for entire images is inappropriate for video systems. Because of the human processing limitation, we will use high resolution information only at the current point of interest. By matching the information content of the image to the information processing capabilities of the human visual system, significant reductions in bandwidth can be realized, provided that the point-of-gaze of the eye is known. Eye gazed compression is a good way to decrease the bandwidth as described in [1] “Implementation of a foveated image coding system for image bandwidth reduction” (Philip Kortum and Wilson Geisler). The reference [1] describes a method using superpixels to reduce the bandwidth of a 256x256 grayscale picture. This algorithm is based on eye gazed compression and the compression changes dynamically. Our approach is different. The reference [1] proposes to have different levels of compression in a same image, which is really efficient but linked to the perception of only one person. We propose a system based on dynamic compression for a display that includes several screens. If several persons watch a display, full global perception, full color images and large images are requested for surveillance purpose. This need of accuracy forbids the use of usual compression methods based on picture interpolation [7][27][28][29]. We have developed a system which matches the human vision area and provide displays equal to the size of this area. The display’s resolution changes according to the eye movement. Finally, a face-based compression allows us to significantly reduce the network traffic by compressing the background and keeping the faces in high resolution.

2 Network Bandwidth Computation

Similar to a digital still picture camera, a network camera captures individual images and compresses them into a JPEG format. The network camera can capture and compress, for example, 30 such individual images per second (30 fps), and then make them available as a continuous flow of images over a network to a viewing station. At a frame rate of about 16 fps and above, the viewer will perceive full motion video (MJPEG). As each individual image is a complete JPEG compressed image, they will all have the same guaranteed quality, determined by the compression and resolution level as defined for the network camera or network video server. For a traditional video camera system, the user chooses in an administration

64

window the frame rate, the compression and the resolution size; these parameters remaining fixed during the use of the system. These parameters are chosen according to the performance of the network. Our objective is to adapt dynamically the parameters to optimize the visualization and the bandwidth use. Traditionally, the screen of interest is displayed in a large window and with several small windows bordering the largest one. Each screen could be displayed following given rules such as event switching or time switching for these compression and resolution parameters would remain the same. In practice, a video camera displayed on the small screen with a high resolution does not provide a significant improvement for the viewer and is a waste of bandwidth. The correlation between image size, compression and resolution level is illustrated schematically in Fig. 1. The size of the picture from the motion JPEG flow increases with the resolution and the quality (less compression).

1030

5070

90160*120

320*240

640*4800

10

20

30

40

Image Size (Ko)

Compression Rate (%)Image

resolution

Image Size function of compression and resolution

30-4020-3010-200-10

Fig. 1. Image size Diagram AXIS-210

We create a model [27] of this function in order to optimize the link between human perception and image size. We obtained three equations (1)(2)(3) to compute the bandwidth. The exponential equation (2) is linked to the nature of JPEG compression. JPEG is designed for compressing full-color or gray-scale images of natural, real-world scenes. JPEG removes redundant information from images which is why equation (2) is based on the exponential: the non-redundant information could not be totally removed and remains as the lowest size of the picture. The image resolution I in pixels and the compression rate percentage CR are the main parameters used to compute the image size S in pixels and the resulting bandwidth B (Mbits per second). First, the non-redundant information RI can be approximated by equation (1). 7.41*410*3.44 +−= IRI (1)

40.0)(* −= CRRIS (2)

SCFPSCNSB ***= (3)

The constant SC represents the complexity of a scene. This constant is in most instances equal to 8. It is used along with the number of cameras CN and the frames per seconds fps for each camera to carefully estimate the bandwidth. Typically, for a set up of four cameras capturing 640*480 video at 30 fps with a compression rate of 10%, the bandwidth required is 36.4 Mbit/sec. This is the reason why a system with more than 12 cameras overtakes the capacity of traditional Ethernet cable (100 Mbits/sec). Equation (3) will be used later to compute the bandwidth for a given image size.

65

3 Human Bandwidth Modeling

In the next section, we will answer three questions: What is the area accurately covered by our view? How much visual information can a human process? How does the human vision behave outside of this small area? The goal is to provide a video surveillance system which matches perfectly the capacity of the human vision (also called the human bandwidth). 3.1 Details perception Human vision is divided in two different processes [29]: a conscious and a subconscious process (Fig. 2). If visual processing did not occur on a subconscious level then the act of seeing would become a labored, arduous and inefficient burden. The peripheral visual processing system is particularly subconscious. This division between the central, conscious role of vision and the unconscious peripheral role is especially important for the field of orientation and mobility.

Fig.2. Two different levels of vision: conscious and subconscious [29]

The conscious vision includes the fovea zone and the attentive vision zone (Fig. 2). The high vision of the fovea zone corresponds to an angle of vision of approximately 3°. Vision is considered to be poor except in this zone. The observer directs the glance by a perpetual movement of the eye to direct the fovea axis towards the part of the image retained for a fine analysis [2]. The close zone constitutes a zone of monitoring whose interpretation allows the fast orientation of the eye towards detail chosen instinctively in spite of a low acuity and without movement of the head. This research relates to certain details of the image which require an intellectual act of interpretation. Table 1. The vision angles, parameters for the Gaussian human distribution [30]

Category Horizontal angle

Vertical up angle

Vertical down angle

Fovea zone +3° +3° -3° Attentive vision +15° +8° -12°

Perceptive vision +50° +35° -50° Lateral vision +100° +50° -75°

To conclude this biological explanation of the human vision field, a normal person can not see any details outside of the fovea zone. Which part of the screen can be seen in full detail? How much information is included in this zone? Table 1 provides some measures of the vision area. 3.2 Limit to Human Vision & its Effect on Optimum Digital Image Resolution: In general, the higher the resolution is, the better the image rendering is. The image would appear cleaner and less jagged. The closer the dots in an image are, the more likely we can only see continuous patterns and not see any dots. The eyes just blur everything together. The human eye only has a certain number of light detectors in it. It is these sensors that convert the light into nerve impulses. Since these sensors are in limited number, it makes sense that they can only handle a certain amount of information. When they have taken in all they can, the brain goes to work and interprets the signals from the sensors, and also determines what is likely to be between the sensors as well. The size of the peak of the cone does not change much, and the eye is constantly working to focus the light on this area. In this area called the “fovea”, the eyes have a densely packed bunch of sensors, which make up the central field of vision [2]. The further away the screen is, the smaller the area it will cover at the end of the cone where it is located. Therefore, when an

66

object is viewed up-close, the eye's central field of vision is filled with the object. A lot of information about the object is sent to the brain, and the detail of the object can be recognized. We will consider a viewer 30 centimeters away from the screen. According to table 1 and some basic geometry, the total area seen is equal to a large video screen (84cm diagonal). In the following demonstration, we will use this as our screen size S. This screen will have a resolution of 1920x1080 pixels.

Fig. 3. Gaussian distribution of the human vision, (left) based on a uniformed distribution of the vertical up angle [30], (right) based on a uniformed distribution of the vertical down angle [30] The human eye distribution can be surrounded by two Gaussian distribution densities ( )11,σmG

and ( )22 ,σmG because of the difference between the up and down vision The human distribution is not uniforme, the vertical down angle value is 5 to 30% bigger than the vertical up angle value (Table 1). For example, the attentive vision has an horizontal angle of 15°, vertical up of +8° and vertical down of -12°. The parameters 1σ and 2σ are estimated from the cone of perception described in the Table 1. This distribution can be considered as a filter for browsing the screen on the area of global vision A. The value Z represents the percentage of the screen (described previously) seen at a distance of 30 cm. It is the result from the density function αf in convolution with the area viewed H at the point ( )., yxM

∫∫⎟⎟⎠

⎞⎜⎜⎝

⎛−

=A

r

dseyxf2

2

21

2 21),( ασ

αα πσ

(4)

S

yxHyxfZ ),(*),(αα = (5)

21 ZZZ ≤≤ (6)

The computation of %211 =Z and %252 =Z show that human bandwidth has a small and limited capacity. Only 25% of an 80cm diagonal screen is seen globally at a distance of 30cm. This is the reason why the information sent to the viewer must be equal to his capacity and must be updated as the views change. For a display that includes hundreds of cameras, the number of screens seen is usually over estimated and network resources are wasted. Understanding the human behavior and capacity allows us to create application which matches these characteristics and the network bandwidth. 3.3 Motion perception Motion perception is an important aspect of the system, especially in video surveillance. It must be enhanced to increase the efficiency of the global system. The peripheral retinal system is sometimes called the "where" retina. It is involved with the subconscious control of human navigation. It is an old visual system, having evolved long before central visual processing. The extreme far edges of the retina are purely reflexive. When an object moves on the far retinal edge an immediate reflex swings the eyes in a direction which aligns the moving object with the fovea. Closer in, the peripheral retinal tissue can "see" movement, but there is no object recognition. When movement stops, the object becomes invisible.

67

According to this biological demonstration, the frame rate can not be reduced without a loss in the level of the perception. We propose a system which will reduce the resolution for non-seen area and keep a high frame rate for viewing comfort.

4 Bandwidth adaptation

4.1 Resolution adaptation To demonstrate the performance of our system, we compare our CMU-system to existing systems. We have established some real scenarios to analyze the bandwidth adaptation when switching between different views. First, the user is watching a display with 4 equal screens. The screen selected is displayed in a full screen mode for 10 seconds. Then the user is returned to the four screen view. This scenario (Fig. 4) is tested with default static parameters: resolution (640x480) and compression (10%). For the dynamic management, these two parameters will change according to the human behavior. As explained previously, the number of frames per second is maintained at 30 to guarantee a full movement perception.

Fig. 4. 4 screens scenario, screen switching every 10 seconds

The first comparative test is based on static bandwidth management. The default resolution (640x480) and compression (10%) are selected permanently. These parameters are chosen in order to obtain high definition video when the user selects a screen in a full screen mode. The results (Fig. 5) were predictable. Because none of the parameters evolve, the bandwidth does not change and 38Mbt/s is used permanently to display 4 cameras in high definition.

Fig. 5. 38Mbt/s used for a static management (100%=100Mbt/s)

The second test is based on basic dynamic management of the bandwidth. When the user switches to the full screen mode, the bit flow from the three other video cameras is disabled. It results a rectangle signal (Fig. 6) with the high level corresponding to the four screen mode (38Mbt/s) and the low level to the full screen mode (9Mbts). However, the bandwidth used globally is still too large to support over 12 video cameras.

Fig. 6. 38Mbt/s – 9Mbt/s for a basic dynamic bandwidth management (100%=100Mbt/s)

68

In the third trial, we develop a full dynamic bandwidth management system in order to adjust the human bandwidth to the network bandwidth. When the watcher switches to a 4 screens display, the image resolution becomes smaller so that the image size diminishes according to the equation (3). As explained previously, the viewer does not notice a change for over a certain resolution. This is why the low resolution fits perfectly the small display. Globally the bit rate for the four screen display decreases from 38Mbt/s to 22Mbt/s (Fig. 7). The gain is approximately 43%.

Fig. 7. 22Mbt/s – 8Mbt/s for an evolved dynamic bandwidth management

4.2 Human Control The bandwidth used for a one screen display can’t be reduced without decreasing the quality of the video. The ratio between video resolution and video quality is optimal. In section 3, the human bandwidth is presented as a limited resource, less than 25% of a screen being noticed that is the reason why it is useless and expensive for a system to provide full detail quality for screens which are not watched. While using this human property, we provide a system that increases the quality of the images viewed at the expense of the quality of the other videos. An eye gazing device can locate the screen watched and transmit the coordinates of the screen observed with an accuracy of a few centimeters. As explained in the chapter 3.3, human attention is sensitive to motion. This is why the frame rate must remain maximal even for the screen with low resolution and high compression.

Fig. 8. Real Time Eye Gazing Fig. 9. Quality On Demand for a 4

screens video display

69

Fig. 10. Flow-chart Eye-gazing bandwidth control

One drawback that was observed in the first versions of the system was that the switching was too frequent because of the eye’s reflex. For example, when screen 1 was watched just after screen 2 the modifications that occurred on screen 2 is perceived as a movement and the eyes focus directly on the previous screen. To avoid this phenomenon, a timer is incorporated to avoid this window wiper effect. To fully understand how this drawback is decimated, the diagram (Fig. 10) describes the dynamic steps of the program. When using the eye gazing, the bandwidth reduction is efficient. Globally, the reduction is over 50%. For our last test system the bandwidth was about 22Mbt/s before the human control and 10Mbt/s after. In the Fig. 11, we can see the difference between the evolved bandwidth management (22Mbt/s) and the human control bandwidth management.

Fig. 11. 22Mbt/s – 10Mbt/s CMU dynamic bandwidth management (100%=100Mbt/s)

5 Face Based Image Compression

At this level, only screen, that are viewed are displayed in high resolution. Can we still reduce the size of the picture while keeping the important information? In the field of video surveillance, it is interesting to capture the face of the people moving in front of the video cameras and while the background is less interesting but needed. We develop a video system including high definition faces and low resolution backgrounds. First, a classifier is trained with a few hundred sample views of faces, which are called positive examples. These are all scaled to the same size. Negative examples are arbitrary images of the same size. After a classifier is trained, it can be applied to a region of interest (of the same size as used

70

during the training) in an input image. The classifier output is "1" if the region is likely to show the face and "0" otherwise. To search for the object in the whole image one can move the search window across the image and check every location using the classifier. The classifier is designed so that it can be easily "resized" in order to be able to find the objects of interest at different sizes, which is more efficient than resizing the image itself. So, to find an object of an unknown size in the image the scan procedure should be done several times at different scales.

Fig. 12. Face based compression method overview

In comparison to the high resolution image (12.1KB) the composite image size is only 4.9KB. The image reduction is about 60%, but the information is fully transmitted. The high definition face is added to the low revolution background to preserve the global quality of the picture. When transmitting the coordinates of the face found in the high definition picture, theses coordinates must be translated into the new referential coordinates of the composite image. Equation (3) is used to compute the bandwidth for face-based compression video. To conclude, we will evaluate all the techniques described in this paper (Table 2).

Table 2. Bandwidth management comparative results Management Full screen 4 screen Bandwidth

average Reduction

Static 38Mbt/s 38Mbt/s 38Mbt/s 0%

Basic Dynamic 9Mbt/s 38Mbt/s 24Mbt/s 39%

Evolved 9Mbt/s 22Mbt/s 16Mbt/s 57%

Human control + Evolved

9Mbt/s 10Mbt/s 9,5Mbt/s 75%

Face based + Human control

+Evolved

4Mbt/s 5Mbt/s 4,5Mbt/s 88%

71

6 Conclusion

The combination of the dynamic bandwidth management for the displays, the eye gazing control and the face-based compression allow for a reduction in bandwidth of up to 88%. This value increases with the number of cameras displayed. However, this system has limitations. The face-based compression is efficient only if faces are detected on the video capture. This systems request a camera with several resolution outputs to compute the composite image which is perceptually close from the original high definition image.

7 Acknowledgement

We would like to thank Army Research Office (ARO) and Bombardier Transportation, USA for their sponsorship. We are also in debt to Mr. Joshua Emerson and Mr. Lawrence R. Gallagher, director of Data Communication for their professional support.

8 References

1 Implementation of a foveated image coding system for image bandwidth reduction, Philip Kortum and Wilson

Geisler, University of Texas Center for Vision and Image Sciences. Austin, Texas 78712 2 Visual Performance. In Bass, M. (Ed.) Geisler, W.S. and Banks, M.S. (1995), Handbook of Optics Volume 1:

Fundamentals, Techniques and Design, 2nd Edition, New York 3 Patent EP20030767118 Turner r Brough (us); Bruemmer Kevin j (us); Matatia Michael(us), methods and

apparatus for network signal aggregation and bandwidth reduction 4 Juday, R.D. and Fisher, T.E. (1989) Geometric Transformations for video compression and human teleoperator

display. SPIE Proceedings: Optical Pattern, Recognition, Vol. 1053, 116-123. 5 Warner, H.D., Serfoss, G.L. and Hubbard, D.C. (1993) Effects of Area-of-Interest Display Characteristics on

Visual Search Performance and Head Movements in Simulated Low-Level Flight. AL-TR-1993-0023. Armstrong Laboratory, Human Resources Directorate, Aircrew Training Division, Williams AFB, AZ.

6 Wassel, H., Grünert, U., Röhrenbeck, J., and Boycott, B.B. (1990) Retinal ganglion cell density and cortical magnification factor in the primate. Vision Research, 30, 1897-1911.

7 Weiman, C.F.R. (1990) Video Compression Via Log Polar Mapping. SPIE, Proceedings : Real Time Image Processing II, Vol. 1295, 266-277.

8 Wilson, H.R., Levi. D., Maffei, L., Rovamo, J. and Devalois, R. (1990). The Perception of Form: Retina to Striate Cortex. In L.S. & J.S. Werner (Eds.), Visual Perception: The Neurophysiological Foundations (pp 232-272). San Diego: Academic Press.

9 Patent US20010865485 Peck Charles c (us); Mackay John d (us) Eye gaze control of dynamic information presentation

10 R. Guerin, H. Ahmadi, and M. Naghshineh, "Equivalent capacity and its application to bandwidth allocation in high-speed networks," IEEE J. Select. Areas Commun., vol. 9, pp. 968-981, 1991. Equivalent capacity and its application to bandwidth allocation inhigh-speed networks

11 R. J. Gibbens , P. J. Hunt, Effective bandwidths for the multi-type UAS channel, Queueing Systems: Theory and Applications, v.9 n.1-2, p.17-28, Oct. 1991

12 M. Sidi, W. Z. Liu, L Cidon, and I. Gopal, "Congestion control through input rate regulation," in Proc. GLOBECOM '89, Dallas, TX, pp. 1764-1768.

13 Jon Kleinberg , Yuval Rabani , Éva Tardos, Allocating bandwidth for bursty connections, Proceedings of the twenty-ninth annual ACM symposium on Theory of computing, p.664-673, May 04-06, 1997, El Paso, Texas, United States

14 San-Qi Li , Song Chong , Chia-Lin Hwang, Link capacity allocation and network control by filtered input rate in high-speed networks, IEEE/ACM Transactions on Networking (TON), v.3 n.1, p.10-25, Feb. 1995

15 Errin W. Fulp , Douglas S. Reeves, Bandwidth provisioning and pricing for networks with multiple classes of service, Computer Networks: The International Journal of Computer and Telecommunications Networking, v.46 n.1, p.41-52, 16 September 2004

16 MeasuringBandwidth K Lai, M Baker - INFOCOM, 1999 - gunpowder.stanford.edu 17 Anwar I. Elwalid Debasis Mitra Effective bandwidth of general Markovian traffic sources and admission control

of high speed networks June 1993

72

18 real-time foveated multiresolution system for low-bandwidth video communication WS Geisler, JS Perry - Proc. SPIE, 1998 - svi.cps.utexas.edu

19 Patent US20040978903 Zimmerman Ofer (il); Stanwood Kenneth l (us); Bourlas Yair (us) Method and apparatus for bandwidth request/grant protocols in a wireless communication system

20 Patent CA20032494956 Kandhadai Ananthapadmanabhan a (us); Manjunath Sharath (in); Bandwidth-adaptive quantization

21 Patent WO2004IB51573 Sethuraman Ramanathan (nl), Method and apparatus for scalable signal processing 22 Patent WO2003EP10523 Riedel Michael (de); Neumann Roland (de), System and method for lossless reduction of

bandwidth of a data stream transmitted via a digital multimedia link 23 Patent US20000633217 Zahorjan John (us); Eager Derek l (ca); Vernon Mary k (us)

Bandwidth reduction of on-demand streaming data using flexible merger hierarchies 24 Patent CN19980807431 Wallerius John w (us); Walters Andrews j (us); Vastano John (us)

Method and apparatus for wireless communication employing control for confidence metric bandwidth reduction 25 Patent WO2003JP1 Kuroki Kenichiro (jp); Tajima Yuji (jp), Image processing display device and image

processing display method 26 Patent US20000568196 Bell Cynthia s (us), Microdisplay with eye gaze detection 27 Patent CA20032496389 Westphal Geoffrey a (us), System and method for image compression, storage, and

retrieva 28 Patent US20030684100 Walker Bradley k (us); Turnipseed John d (us); Method and apparatus for providing

interactive multimedia and high definition video 29 Institute for innovative blind navigation http://www.wayfinding.net/vsionsys.htm 30 http://www.rennes.supelec.fr