41
1 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu Social Interactions over Geographic-Aware Multimedia Systems ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, 2013 Roger Zimmermann & Yi Yu Module 4 Media Development Authority Media Management Research Lab School of Computing National University of Singapore

ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

1

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Social Interactions over Geographic-Aware Multimedia

Systems

ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, 2013

Roger Zimmermann & Yi Yu

Module 4

Media Development Authority

Media Management Research Lab School of Computing

National University of Singapore

Page 2: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

2

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Outline: Module Four

I.  GeoVid  Resources    II.  MPEG-­‐DASH  Overview  and  Resources  III. Google  Street  View  Hyperlapse  IV. SyntheBc  Video  Meta-­‐Data  

GeneraBon  V.  Conclusions  

2

Page 3: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

3

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

GeoVid  Architecture  (1)  

Page 4: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

4

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

GeoVid  Architecture  (2)  

•  Sensor-rich video management design – Upload interface – Fast content availability – Spatio-temporal query support – RESTful GeoVid APIs – Support for HTML5 video streaming standard – Ready for MPEG-DASH video streaming standard – Mobile apps (iOS and Android)

Page 5: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

5

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

GeoVid  API  (1)  •  URL:  hKp://api.geovid.org  

Page 6: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

6

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

GeoVid  API  (2)  •  GeoVid  API  DescripBons  

Page 7: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

7

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

GeoVid  API  (3)  •  Ex.:  Search  Service  

Page 8: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

8

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Outline: Module Four

I.  GeoVid  Resources    II.  MPEG-­‐DASH  Overview  and  Resources  III. Google  Street  View  Hyperlapse  IV. SyntheBc  Video  Meta-­‐Data  

GeneraBon  V.  Conclusions  

8

Page 9: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

9

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu      

Dynamic  AdapBve  Streaming  over  HTTP  (DASH)  

•  Traditional media streaming (e.g., based on RTP/RTSP/RTCP) streaming faces several challenges in large-scale systems: –  Special-purpose servers/proxies/peers for media

(complex) –  Difficulties in firewall traversal: protocols use both TCP

and UDP transmissions –  Difficulties in caching data (no “web caching”)

•  Advantage of RTP/RTSP/RTCP –  Short end-to-end latency

Page 10: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

10

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu      

DASH  Standard  (Dec.  2011)  •  ISO/IEC Standard:

–  “Information technology — MPEG systems technologies — Part 6: Dynamic adaptive streaming over HTTP (DASH)”

–  JTC 1/SC 29; FCD 23001-6 •  Uses HTTP protocol to “stream” media •  Divides media into small chunks, i.e., streamlets •  Makes playback adaptive

–  Encode media into multiple different streamlet files, e.g., a low, medium, and high quality version (different bandwidth).

•  Supported by: Move Networks, Apple, Microsoft, Netflix, …

Page 11: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

11

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

     

DASH  LimitaBons  

•  DASH standard only describes server-to-client dissemination: – Upload mechanism is undefined!

•  We are working on a low latency DASH upload protocol for mobile devices.

MPD:  Media  PresentaBon  DescripBon  

Page 12: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

12

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

DASH  Upload  Design  

Page 13: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

13

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Source  Video  Stream  

•  720p  

CharacterisBcs  of  the  source  video  stream  

¢  480p  

Page 14: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

14

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

EvaluaBon  Metrics   startupT

playerserver

T

uploadsegstartup TTTTTuploader

+++≈!"!#$

Different delay components and their relationships.

Page 15: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

15

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

SegmentaBon  Delay  

Normalized median segmentation time (processing time/segment duration).

Values less than 1 indicate that a device is capable of creating a segment within playback duration, which means continuous processing of long videos in a pipelined manner is possible.

segT

¢  720p   ¢  480p  

Segment  duraBon:  3s   Segment  duraBon:  10s  

Page 16: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

16

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

SegmentaBon  Overhead    

¢  720p  iPhone 4

Normalized segmentation delay of 720p video on iPhone 4.

Real-time segmentation of high-quality video will be feasible in any mobile platform in the future.

Page 17: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

17

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

WiFi  Transmission  Delay   uploaderT

Network delivery of 480p is quite possible in a stable wireless environment, while 720p video, even with small segment duration, experience network bottleneck.

<1   >1  

Page 18: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

18

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

MPEG-­‐DASH  Resources  •  Good introduction and overview:

–  “Dynamic Adaptive Streaming over HTTP: Standards and Design Principles,” Thomas Stockhammer, ACM Multimedia Systems, 2011.

•  Dataset: –  “Dynamic Adaptive Streaming over HTTP Dataset,” S.

Lederer, C. Müller, C. Timmerer, ACM Multimedia Systems, 2012.

•  ACM MMSys web site: http://www.mmsys.org/ •  DASH-JS

–  An open source reference DASH player written in JavaScript from the DASH Industry Forum.

Page 19: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

19

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Outline: Module Four

I.  GeoVid  Resources    II.  MPEG-­‐DASH  Overview  and  Resources  III. Google  Street  View  Hyperlapse  IV. SyntheBc  Video  Meta-­‐Data  

GeneraBon  V.  Conclusions  

19

Page 20: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

20

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Google  Street  View  Hyperlapse  

•  Google Street View consists of geo-referenced, panoramic images.

•  Project: Creating a time-lapse video from Google Street View images. –  JavaScript hyper-lapse utility for Google Street View. –  http://hyperlapse.tllabs.io/

Page 21: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

21

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Hyperlapse  Demo  Video  

•  http://vimeo.com/63653873

Page 22: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

22

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Outline: Module Four

I.  GeoVid  Resources    II.  MPEG-­‐DASH  Overview  and  Resources  III. Google  Street  View  Hyperlapse  IV. SyntheBc  Video  Meta-­‐Data  

GeneraBon  V.  Conclusions  

22

Page 23: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

23

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Ä “Capture the sensor inputs and fuse them with the video streams” §  Recorded 134 video clips using the recording prototype system in Moscow, ID (total 170 mins video). §  Videos covered a 6km by 5km region quite uniformly. §  Average camera movement speed was 27km/h, and average camera rotation was around 12 degrees/s. §  Collected meta-data included 10,652 FOV scenes in total.

Real-­‐World  Video  CollecBon  

Page 24: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

24

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Ä The collected real-world video data has not been large enough to evaluate realistic applications on the large scale. §  Collecting real-world data requires considerable

time and effort. Ä A complementary solution is to synthetically generate

georeferenced video meta-data.

Real-­‐World  Video  CollecBon  Challenges  

Page 25: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

25

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Output:  Georeferenced  Video  Meta-­‐data      

Calculate    Moving  Direc4on  

Adjust  Direc4ons  on  Turns   Randomize  Direc4on  Angles  

Camera  direc4on  computa4on  

The  Brinkhoff  Algorithm   The  GSTD  Algorithm  

Merge  Trajectories  

Input:    Camera  Template    SpecificaBon  

Camera  movement  computa4on  TIGER/Line  

Files  

network-­‐based  movement    

free  movement  

mixed    movement    

SyntheBc  Video  Meta-­‐data  GeneraBon  

Page 26: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

26

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

26  

Ä  Adopted the Brinkhoff algorithm for camera trajectory generation

Network-based Movement

Ä  Introduced stops and acceleration/ deceleration events at some road crossings and transitions

npnBvv prevnext),(

∗=

Ä  In a deceleration event reduction in camera speed is simulated based on the Binomial distribution

When n=20 and p=0.5 speed is reduced to half at every time instant

Ä  Camera accelerates with a constant rate (user defines the acceleration rate)

Ä  Cameras move on a road-network

Camera  Movement  ComputaBon  (1)  

Page 27: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

27

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Ä  Improved the GSTD algorithm to generate the camera trajectories with unconstrained movement: p  Added speed control mechanism p  Camera movement data is generated in geographic coordinate system

(i.e., as latitude/longitude coordinates)

Free Camera Movement

Ä Cameras move freely.

Camera  Movement  ComputaBon  (2)  

Page 28: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

28

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Mixed Camera Movement Ä  Cameras sometimes follow the network and sometimes move randomly

on an unconstraint path.

i.  Generate a network based trajectory (Tinit)

ii.  Randomly select n sub-segments (S1 through Sn) on  the  trajectory      0≤|Si|  ≤  (Tinit/4)  and  Nrand=  (∑Si)  /  |Tinit|)  (user  defines  Nrand)  

i.  Replace Si with Trand(i) ii.  Update timestamps

Camera  Movement  CompuaBon  (3)  

Page 29: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

29

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Ä  Assigning meaningful camera direction angles is one of the novel features of the proposed data generator.

Output: Georeferenced Video Meta-data

Calculate    Moving  Direc4on  

Adjust  Direc4ons  on  Turns   Randomize  Direc4on  Angles  

Camera  direc4on  computa4on  

Fixed camera:

1) Calculate moving direction

2) Adjust directions on turns

Random rotation camera:

1)  Calculate  moving  direcBon    

3)  Randomize  direcBon  angles  

2)  Adjust  direcBons  on  turns  

Camera  RotaBon  CompuaBon  (1)  

Page 30: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

30

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Fixed Camera

1) Calculate moving direction 2) Adjust directions on turns

Rotation angle from t1 to t2 is larger than ωmax (i.e., rotation threshold)

Smooth down the rotation by distributing the rotation amount forwards and backwards

Trajectory Tk

is the moving direction vector at time t

t1

t2

Camera  RotaBon  ComputaBon  (2)    

Page 31: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

31

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Fixed Camera

Ä  Illustration of camera direction adjustment for vehicle cameras

Real-world data Synthetic data before direction adjustment

Synthetic data after direction adjustment

Camera  RotaBon  ComputaBon  (3)  

Page 32: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

32

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Randomly Rotating Camera

1) Calculate moving direction

3) Randomize direction angles

2) Adjust directions on turns

p  Randomly rotate the directions at each sample point towards left or right

p  Rotation amount is inversely proportional to the current camera speed level

p  The rotation amount is guaranteed to be less than rotation threshold ωmax

Camera  RotaBon  ComputaBon  (3)  

Page 33: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

33

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Ä  Goal: Evaluate the effectiveness of the synthetic data generation approach through a high level comparison between the real-world and synthetic data.

Ä  Datasets: p  Generated two groups of synthetic data:

1.  Using vehicle camera template 2.  Using passenger camera template

p  Both synthetic data groups were created based on the road network of Moscow, ID.

Ä  Methodology: p  Analyze and compare the movements and rotations of real-world and

synthetic datasets. p  Report:

1.  The average and maximum values for speed and rotation 2.  Frequency distribution of different speed and rotation levels

Experimental  EvaluaBon  (1)  

Page 34: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

34

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Comparison of Camera Movement Speed Maximum  speed  (km/h)   Average  speed  (km/h)   StdDev  of  speed  

Synthe4c  data  with  fixed  camera   87.91   27.14   12.82  

Synthe4c  data  with  free  camera  rota4on   87.28   27.32   13.01  

Real-­‐world  data   0.564   27.03   13.68  

Characteristics of the camera speed

Comparison of camera speed distributions for real-world data and synthetic data with fixed camera Illustration of camera movement speed on map Real-­‐world  data   SyntheBc  data  

Experimental  EvaluaBon  (2)  

Page 35: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

35

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Comparison of Camera Rotation Maximum  rotaBon(degrees/s)   Average  rotaBon(degrees/s)   StdDev  of  rotaBon  

Synthe4c  data  with  fixed  camera   32.33   4.64   7.24  

Synthe4c  data  with  free  camera  rota4on   55.27   12.59   9.35  

Real-­‐world  data   107.30   11.53   14.02  

CharacterisBcs  of  the  camera  rotaBon  (ωmax  =60  degrees)  

Comparison  of  camera  rotaBon  distribuBons  for  real-­‐world  data  and  syntheBc  data  with  fixed  camera   IllustraBon  of  camera  rotaBon    on  map  

Real-­‐world  data   SyntheBc  data  Comparison  of  camera  rotaBon  distribuBons  for  real-­‐world  data  and  syntheBc  data  with  random  rotaBon  camera  

Experimental  EvaluaBon  (3)  

Page 36: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

36

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Performance Issues

Camera  Template   Trajectory  PaKern  

RotaBon  PaKern   Number  of  Videos  

Time  to  Generate  camera  trajectory  (s)  

Time  to  Assign  DirecBons  

Total  Time  

Vehicle  camera   Tnetwork   Fixed  camera   2,980   124   39   163  

Passenger  camera   Tnetwork   Random  rota4on   2,980   115   201   316  

Pedestrian  camera   Tfree   Random  rota4on   2,970   32   263   255  

Pedestrian  camera   Tmixed   Random  rota4on   2,970   271   215   486  

Ä  The generator can create synthetic datasets in a reasonable amount of time with off-the-shelf computational resources.

Ä  The measured data generation times for different types of datasets and parameter settings.

Experimental  EvaluaBon  (4)  

Page 37: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

37

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Ä Proposed a two step synthetic data generation 1.  Computation of the camera movements 2.  Computation of the camera movements

Ä Compared the high-level properties of the synthetically generated data and those of real-world geo-referenced video data.

Ä The synthetic meta-data exhibit equivalent characteristics to the real data, and hence can be used in a variety of mobile video management research.

Summary  

Page 38: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

38

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Conclusions  •  Annotation using sensors can provide automatic and

objective meta-data for indexing and searching. •  Georeferenced video search has a great potential,

especially in searching user generated videos. •  Many open questions:

–  Standard format of meta-data –  Standard way of embedding meta-data –  Index structures of meta-data for fast searching –  Supporting new query types –  Combining with content based features –  Relevance ranking for result presentation

Page 39: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

39

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

Future  Work  

•  Sensor values are sometimes noisy and contain errors.

•  Better alignment of video screen position with mirror world.

•  Support for multiple videos. •  When to rotate the video plane and when to move

the mirror world viewpoint. •  Addition of 3D query facility.

Page 40: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

40

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

©  Yi  Yu  and  Roger  Zimmermann  This research has been supported in part by the Singapore National Research Foundation under its International Research Centre @ Singapore Funding Initiative and administered by the IDM Programme Office. Some of the work has been performed under the Centre of Social Media Innovations for Communities (COSMIC).

Acknowledgment  

Page 41: ACM Multimedia 13, Barcelona, Catalunya, Spain October 21-25, … · 2017-09-17 · 4 ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module

41

ACM Multimedia 2013 Tutorial: Social Interactions over Geographic-Aware Multimedia Systems (Module 4) – Roger Z & Yi Yu

End of Module Four

Thank You – Q&A

hYp://geovid.org  Further  informaBon  at:  

[email protected]  [email protected]