Face Recognition Application Guidance · Face Recognition Application Guidance Overseas Technical Support Team 12 Rule, tick ‘Rule’ and draw the detection area and min. pupil

Face Recognition Application Guidance Overseas Technical Support Team

1

Face Recognition Application Guidance

Change time Contents Editor

20180313 New add [email protected]

20180522 Edit [email protected]


2

Content

Face Recognition Solution Guidance ................................................................................................. 1

Chapter 1. Solution Introduction .............................................................................................. 3

1.1 Solution 1: Camera + VMS ........................................................................................... 3

1.2 Solution 2: Camera + NVR + VMS ................................................................................ 3

1.3 Solution 3: Camera + Server ........................................................................................ 4

1.4 Comparison of three Solutions.................................................................................... 5

Chapter 2. Camera Installation Specification ............................................................................ 6

2.1 Installation Location .................................................................................................... 6

2.2 The Installation Requirement of Camera .................................................................... 7

2.3 The Selection of Camera Lens ..................................................................................... 8

2.4 Installation Position and Lens Query Table ................................................................. 8

Chapter 3. Camera Configuration.............................................................................................. 9

3.1 Configure the parameter via IE browser ..................................................................... 9

Chapter 4. DeepInmind NVR Operation .................................................................................. 10

4.1 Create Database ........................................................................................................ 10

4.2 Configure Face Detection .......................................................................................... 11

4.3 Configure Face Comparison ...................................................................................... 13

4.4 Search Captured face ................................................................................................ 14

Chapter 5. Face Recognition Server Operation ....................................................................... 16

5.1 Setup Face Recognition Server .................................................................................. 16

5.2 Add Cameras to Server .............................................................................................. 18

5.3 Set Face Picture Library ............................................................................................. 19

Chapter 6. Result Show ........................................................................................................... 20

6.1 Solution 1 Show ......................................................................................................... 20

6.2 Solution 2 Show ......................................................................................................... 21

6.3 Solution 3 Show ......................................................................................................... 23

Appendix I. Frequently-used Material Link ............................................................................. 25


3

Chapter 1. Solution Introduction

1.1 Solution 1: Camera + VMS

In this solution, we use face comparison camera to capture human faces and do face comparison.

The results can be displayed by IVMS 4200. The topological diagram is shown as Figure1. This

system is very simple which is very suitable for small projects.

iDS-2CD8426G0/F-I

Face Capture & Face Comparison

iVMS-4200

Face capture Result

Face Comparison

Result

Figure 1 Solution 1 diagram

Models and software version requirement:

Model Version

Camera iDS-2CD8426G0/F-IZS Latest base-line version

VMS IVMS4200 Latest base-line version

Table 1 Solution1 models and software version requirement

1.2 Solution 2: Camera + NVR + VMS

In this solution, we can use different kind of cameras, including not only face capture cameras but

also normal cameras, combine with deepinmind NVR to achieve face comparison function. The

results can be displayed by NVR local GUI or IVMS 4200. The topological diagram is shown as

Figure2.

Face Capture cameras

iVMS-4200

Captured pictures

Normal IP cameras Video

Face comparison

Face Capture & comparison

Figure 2 Solution2 diagram


Recommend Models Version

Camera iDS-2CD8426G0/F-IZS Latest base-line version


4

DS-2CD7126G0-IS

DS-2CD7026G0-IS

DS-2CD7A26G0-IS(H)(Y)

iDS-2PT9122DE/F

NVR iDS-9632NXI-I8/4F

iDS-96000NXI-I16

Latest base-line version

VMS IVMS4200 Latest base-line version


1.3 Solution 3: Camera + Server

In this solution, we use face capture cameras and standalone face recognition server. The camera

capture human faces and send pictures to standalone server. Standalone server can do face

comparison, blacklist alarm, face search, 1V1 comparison. The results can be displayed by

standalone server webpage. The topological diagram is shown as Figure3.

Face capture cameras

Face PictureStand-alone server

Figure 3 Solution3 diagram


Recommend Models Version

Camera iDS-2CD8426G0/F-IZS

DS-2CD7126G0-IS

DS-2CD7026G0-IS

DS-2CD7A26G0-IS(H)(Y)

iDS-2PT9122DE/F

Latest base-line version

Server Latest base-line version



5

1.4 Comparison of three Solutions

Solution Face

Comparison

Speed

Search

by face

picture

Face Library Face Library Picture

Requirements

1 5 pics/sec _

3 Libraries; up to

90,000 pictures;

Face Library

storage: up to

3GB.

40 <Min. Pupil Distance <300

(recommend 60) pixels.

20KB <Photo Size<300KB

JPEG

2

Up to 40

pics/sec support

iDS-9632NXI-

I8/4F: 4 libraries,

up to 100,000

pictures

iDS-96000NXI-

I16: 16 libraries,

up to 300,000

pictures



20KB <Photo Size<4MB

JPG, JPEG

3

Up to 80

pics/sec

support

16 Libraries: up to

1000,000 pictures



20KB <Photo Size<8MB

JPG, JPEG


6

Chapter 2. Camera Installation Specification

In order to have a better face recognition performance, please refer to this chapter for camera

installation.

2.1 Installation Location

The accuracy rate of face recognition is great involved with installation location, ambient light

(such as too dark, too bright) and so on. In order to ensure the better effect, it raises some

installation suggestions as follows:

1) Installation position should be the standard passageways or the Entrances/ Exits, which

insures the direction of traffic is particular and capture the Enter/Leave personnel’ faces in

the direction.

2) Choose the stable lighting condition with adequate illumination. Light compensation is

necessary to ensure the facial features are visible under the condition of insufficient light or

backlight.

Example of standard scenario:

Figure 4 Example of standard scenario

Example of unsuitable scenario:

Figure 5 Example of unsuitable scenario


7

2.2 The Installation Requirement of Camera

Figure 6 The installation diagram

The installation position selection specification of face capture camera is as follows:

1) The camera should be installed in the front of passageway and capture the positive face. The

horizontal angle of deflection had better be less than 25 °.

2) The installation of camera needs an elevation angle, which avoids the situation that the rear

face is obscured when the tandem persons go through the passageway. The vertically

elevation angle should meet the below requirement: α=15±5°.

3) It requires that the covered PD (Pupil Distance) pixels needs to be larger than 40 pixels to

distinguish the face details from the captured image. The practical width of face detection

should be less than or equal to 3 meters with 2MP camera.

4) Make sure that there is no shelter between the camera lens and the passageway.

Example of standard scenario;

Figure 7 Example of standard scenario


8

Example of unsuitable scenario:

Figure 8 Example of unsuitable scenario:

2.3 The Selection of Camera Lens

1) Different cameras, lens focal lengths and monitoring widths are key points for the different

monitoring distance and installation. The conversion relation between them is as follows:

2) The selection of lens focal length： D8.1f 8426 (1)

3) The height of camera：5.1)(tan DH

(2)

4) D : The monitoring distance

5) The height below the head sets as average value 1.5 m

6) α is the elevation angle

7) Elevation angle is recommended as 10°, )10(tan ≈0.18，

)15(tan ≈0.27，

)20(tan

≈0.36

2.4 Installation Position and Lens Query Table

Camera

Model

Monitoring

Width W

Lens Focal

Length

Monitoring

Distance

Installation Height

of Camera

Elevation

angle α°

200W 3.0m 8-32mm 3.3m 2.1m-2.7m 15°±5°

200W 3.0m 8-32mm 6.7m 2.7m-3.9m 15°±5°

200W 3.0m 8-32mm 13.3m 3.9m-6.3m 15°±5°


9

Chapter 3. Camera Configuration

3.1 Configure the parameter via IE browser

Step 1: Enter [configuration]-[local], set the [play performance] as fluent; Enable [Rules]. Then Click

“Save” when finishing the configuration, as shown in the below picture:

Figure 9 Configuration → Local Interface

Step 2: Enter [configuration]-[VCA]-[Rule] interface, check the “Rule”. Then configure the “Min.

Pupil Distance” and “Detection area”. Click “Save” when finishing the configuration.

a) 【Min Pupil Distance】: Minimum size filter box, which is used to control the size of the

captured face. The maximum size box will be generated automatically by default, which the

rectangular box is generated according to the distance between two eyes and mouth. Min

Pupil Distance should be set up by drawing the box instead of entering the number directly.

Moreover it is necessary to set up it, or it cannot capture face picture successfully. Different

image resolutions require different box sizes. With 1080P resolution as an example, the size

value of “Min Pupil Distance” shouldn’t be less than 40, and 60 is recommended.

b) 【Rules】: Draw a rectangle or polygonal (up to 10 sides) face detection area. The minimum

value is 28*28 pixels and the maximum size is to cover the full screen. Detection area is

generally recommended to be drawn as the half of the screen area size, no more than two-

thirds of the picture.

Figure 10 Set Rules

Step 3: 【Advanced Configuration】: The parameters of Advanced Configuration are generally set


10

as default without modification. If want to realize self-comparison or comparison with “Face

Recognition NVR”, “Capture Times” is suggested to modified to 5 times, which increases the

number of uploading face pictures. Click “Save” when finishing the configuration.

Advanced parameter meaning:

【Generation Speed】It is used to control the speed of target generation in the detection region,

the higher the value, the faster the target to generate

【Capture Times】It is used to set up the number of capture times in the detection area during

the process from entering to leaving of face..

【Sensitivity】It is used to control the face detection sensitivity. The higher the sensitivity, the

easier it is to detect the face.

【Capture Interval】It will capture the face image every few frames during the process from

entering to leaving the detection area. The quality of the captured image is highest among these

few frames.

【Capture Sensitivity】It is used to set the score threshold of captured picture. The face image can

be captured only the score of face is greater than or equal to the threshold.

Chapter 4. DeepInmind NVR Operation

If solution 2 is adopted, please refer to this chapter for NVR operation.

4.1 Create Database

Go to NVR’s local GUI, Smart Analysis—Face Picture Database, create database and add face

pictures in the database (The database also can be created via web or iVMS). iDS-9632NXI-I8/4F

can support 4 face libraries in database and each library can support up to 50000 pictures.

Here are the requirements for face picture in database:

40 <Min. Pupil Distance <300 (recommend 60) pixels.

20KB <Photo Size<4MB; Format: JPG, JPEG

Figure 11 Create database via local GUI


11

Figure 12 Create database via web

Notes:

1) Insert the USB drive that contains the face pictures to create database via local GUI. You can

select one picture in USB drive to add or a file that contains multiple pictures to import

pictures in batch.

2) The local GUI don’t support custom tags except name and phone number for database. You

can add custom tags for database via web or iVMS.

3) The picture added to the database should has only one face.

4.2 Configure Face Detection

Face detection function should be enabled if the camera is a normal camera.

Step1. Go to NVR’s local GUI, System—Event—Smart Event—Face detection, tick ‘Enable Face

Detection’ and NVR will detect and capture face by itself.

Figure 13 Enable face detection via NVR GUI

Step2. Then configure rule via NVR’s web page. Go to NVR’s web page, Configuration—VCA—


12

Rule, tick ‘Rule’ and draw the detection area and min. pupil distance (there will be recommended

range of the min. pupil distance below).

Figure 14 Configure rule via NVR’s web page

Figure 15 Advanced parameters of face capture

Notes:

1) Capture Interval: Capture are taken every few frames, in which the captured images are the

highest quality images in each of these frames.

2) Capture Times: To control the number of face capture during the people entering the

detection area and leaving the detection area.

3) Sensitivity: The higher the sensitivity, the easier it is to detect faces.

4) Generation Speed: Target generation speed is used to control the speed of the target

generated by the human face entering the detection area. The larger the value, the faster the target


13

is generated.

5) For face capture camera, the face detection can’t be enabled and the NVR will receive the

face image captured by camera by default.

6) NVR can capture face for the IPC whose resolution is lower than 4MP.

4.3 Configure Face Comparison

Step1. Go to NVR’s local GUI, System—Event—Smart Event—Human Face Comparison, tick

‘Enable Face Picture Comparison’ and select library that you want to compare with. Then set a

value for similarity and select mode.

Step2. There are two modes: Normal and Stranger Comparison. When the Normal mode is

selected, only when the similarity between capture face and face in library is higher than the

Similarity you set, the Human Face Comparison alarm will be triggered. When the Stranger

Comparison mode is selected, only when the similarity between capture face and face in library

is lower than the Similarity you set, the Human Face Comparison alarm will be triggered.

Figure 16 Configure face comparison via NVR GUI


14

Figure 17 Set linkage action via NVR GUI

4.4 Search Captured face

Go to NVR’s local GUI, Smart Analysis—Smart Search—Face Search, there are three kinds of face

search: Search by Event, Search by Picture and Search by Name. In Search by Event, you can

search the capture face by the type of the alarm mode.

In Search by Picture, you can search the capture face by the pictures (When you upload several

pictures, the search results are the combination of these several people) that you upload from

local or Face Picture Database.


15

In Search by Name, you can search the capture face by the name in Face Picture Database.

You can select face picture in the Face Search result and add the selected face picture to Face

Picture Database directly.


16

Chapter 5. Face Recognition Server Operation

If solution 3 is adopted, please refer to this chapter for face recognition server operation.

5.1 Setup Face Recognition Server

Step1. Hardware connect.

Wiring at least 3 twisted pairs:

Connect either LAN A or LAN B Connect LAN 1 and LAN 2 at the same time.

Data Port: TX1 node is for data acquisition，LAN1 and LAN2 use different IP address.

Power Redundant Module: There will be beep with only one power supply.

Step2. Device activate.

Face recognition server management port (Lan A) default IP: 192.168.1.64


17

Note：If you can’t login, please check weather TLS1.1 and TLS1.2 are clicked in the Internet

options

Note ：IP (lanA,lan1,lan2) in same network segment.

Note：Do not tick DHCP

Enter Service>Smart Unit;

Click “Local”, and choose Network;

Set data port LAN1 and LAN2 IP address;

Set management port (LAN A) IP address;

After IP setting is finished, it needs to reload the device using the new IP address you set

Step3. Add to Cluster

Check “Smart Unit”>“Local”;

Click “Create Cluster”;

Refresh the web page, and there will be a cluster IP address showing “activated” at the

top-right corner, which means the cloud analysis activation is successful.


18

5.2 Add Cameras to Server

Step1. Add front resource. Enter Resource> Device;

Step2. Click “Add” on the left side to add organization;

Step3. Click “Administrator-admin” and click “Add” on the right side to add front-end device.

Step4. Click new-built organization (like test), and click “Add” at the right side to add front-end

device to this area;


19

Step5. Click one camera and click “Arm”. Notice: the face recognition server can’t get captured

pictures from the front-end devices without arming.

5.3 Set Face Picture Library

Step1. Enter the blacklist management page;

Step2. Click [+] to add the blacklist library;

Step3. Select the blacklist library, click [add personnel to add], add the blacklist person image;

Step4. Relevant front-end resource channel in the Arming page;


20

Chapter 6. Result Show

6.1 Solution 1 Show

Connect camera to IVMS 4200. Select Face Picture Comparison Alarm module. In this module, we

can see live view, real time face capture and comparison result.


21

6.2 Solution 2 Show

DeepInmind NVR comparison result show.

NVR live view interface:

IVMS 4200 interface:


22

Face search from NVR local GUI:

DeepInmind NVR support face retrieval from IVMS 4200. We can import a face picture and search

similar face pictures from captured face.


23

6.3 Solution 3 Show

From the webpage of standalone server, we can get live view and real time face capture and

blacklist alarm.

Search captured pictures from webpage:

1V1 comparison:


24


25

Appendix I. Frequently-used Material Link

1. Deeplearning promotion package：

ftp://hikftp.hikvision.com:400/Product%20File(Overseas)/00 Oversea Products/01 IP/00

Promotion Package/15 DeepinView/


Promotion Package/14 DeepinMind/

2. DeepInView spec：

ftp://hikftp.hikvision.com:400/Product%20File(Overseas)/00 Oversea Products/01 IP/01 SPEC,

A&E SPEC/SPEC/01 IPC/18 DeepinView/


A&E SPEC/SPEC/02 PTZ/8 DeepinView/

3. Deep Inmind spec：


A&E SPEC/SPEC/03 NVR/DeepinMind NVR/

4. Training material：

ftp://hikftp.hikvision.com:400/Product%20File(Overseas)/00 Oversea Products/10

Training/Training (AI)/

5. AI solution：Guidance


Installation&How to/04 AI/03 Solution/

ftp://hikftp.hikvision.com:400/Product File(Overseas)/00 Oversea Products/01 IP/00 Promotion Package/14 DeepinMind/

ftp://hikftp.hikvision.com:400/Product File(Overseas)/00 Oversea Products/01 IP/00 Promotion Package/14 DeepinMind/

Documents

Face Recognition Application Guidance · Face Recognition Application Guidance Overseas Technical Support Team 12 Rule, tick ‘Rule’ and draw the detection area and min. pupil