|
US$329.00 · In stock
Delivery: <= 3 days. True-PDF full-copy in English will be manually translated and delivered via email.
GA/T 1344-2016: Face recognition application in security systems. Technical requirements for face image extraction from video
Status: Valid
| Standard ID | Contents [version] | USD | STEP2 | [PDF] delivered in | Standard Title (Description) | Status | PDF |
| GA/T 1344-2016 | English | 329 |
Add to Cart
|
3 days [Need to translate]
|
Face recognition application in security systems. Technical requirements for face image extraction from video
| Valid |
GA/T 1344-2016
|
PDF similar to GA/T 1344-2016
Basic data | Standard ID | GA/T 1344-2016 (GA/T1344-2016) | | Description (Translated English) | Face recognition application in security systems. Technical requirements for face image extraction from video | | Sector / Industry | Public Security (Police) Industry Standard (Recommended) | | Classification of Chinese Standard | A91 | | Classification of International Standard | 13.310 | | Word Count Estimation | 14,122 | | Date of Issue | 2016-09-14 | | Date of Implementation | 2016-09-14 | | Quoted Standard | GA/T 893-2010 | | Issuing agency(ies) | Ministry of Public Security | | Summary | This standard specifies the technical requirements and test methods for video face extraction in secure face recognition applications. This standard applies to the security face recognition application in the video face image extraction program design, related product development and testing. | GA/T 1344-2016: Face recognition application in security systems. Technical requirements for face image extraction from video
---This is a DRAFT version for illustration, not a final translation. Full copy of true-PDF in English version (including equations, symbols, images, flow-chart, tables, and figures etc.) will be manually/carefully translated upon your order.
Face recognition application in security systems.Technical requirements for face image extraction from video
ICS 13.310
A91
People's Republic of China Public Safety Industry Standard
Security facial recognition application
Video facial image extraction technical requirements
Released on.2016-09-14
2016-09-14 Implementation
Issued by the Ministry of Public Security of the People's Republic of China
Table of contents
Foreword Ⅰ
1 Scope 1
2 Normative references 1
3 Terms and definitions 1
4 Composition 3
5 Technical requirements 3
5.1 Functional requirements 3
5.2 Performance requirements 3
6 Test method 3
6.1 Overview 3
6.2 Function test 4
6.3 Performance test 6
6.4 Test file 8
Appendix A (Normative Appendix) Standard Video Requirements for Video Face Image Extraction Test 9
Appendix B (informative appendix) Test file record format 11
Security facial recognition application
Video facial image extraction technical requirements
1 Scope
This standard specifies the technical requirements and test methods for video facial image extraction in security face recognition applications.
This standard applies to the design of the video face image extraction scheme, the development and testing of related products in the application of security face recognition.
2 Normative references
The following documents are indispensable for the application of this document. For dated reference documents, only the dated version applies to this article
Pieces. For undated references, the latest version (including all amendments) applies to this document.
GA/T 893-2010 Application terminology of security biometric identification
3 Terms and definitions
The following terms and definitions defined in GA/T 893-2010 apply to this document.
3.1
Video face image extraction faceimageextractionfromvideo
The process of discovering and obtaining facial images from video image sequences as required.
3.2
Face detection
The process of finding a face image from a given video image sequence and determining its position and size.
3.3
Face clustering
The process of grouping and identifying face images according to the number of people appearing in the video.
3.4
Face selection faceselection
The process of selecting an image that meets the predetermined requirements from the classified and grouped face images.
Note. The predetermined requirements include the number of pixels between the eyes, the angle of the face pose, expressions and accessories, etc.
3.5
Ratiooffalsedetection
The proportion of non-face images in the face detection results to the total number of detected images. which is.
Rfd=
Bf
B ×100%
(1)
Where.
Rfd---false detection rate, %;
Bf---the number of non-face images;
B --- Total number of detected images.
Dm---The number of times that a person who meets the predetermined requirements for a face image is not selected;
D ---The total number of people should be selected.
4 composition
Video face image extraction consists of face detection, face clustering and face selection.
First, perform face detection on the video image sequence, and then perform face clustering on the results to form a classification group, and then analyze each group of face images
Make face selection. Face detection is mandatory. Face clustering and face selection are optional. Face selection should be done after face clustering is turned on
get on.
5 Technical requirements
5.1 Functional requirements
5.1.1 Face detection
Should have the following functions.
a) Set the face detection area in the screen, and only detect the face image in the area;
b) Detect the face images of multiple people appearing at the same time;
c) Set the detection range of face size, and only detect faces within the range.
5.1.2 Face clustering
Should have opening and closing functions.
5.1.3 Face selection
Should have the following functions.
a) Turn on and off;
b) Set the upper limit of the number of face images selected per person;
c) Mandatory selection of face images that do not meet predetermined requirements.
5.2 Performance requirements
5.2.1 Face detection
When the false detection rate is not more than 1%, the missed detection rate should not be more than 1%.
5.2.2 Face clustering
When the mispolymerization rate is not more than 5%, the mispolymerization rate should not be more than 10%.
5.2.3 Face selection
When the false selection rate is not more than 5%, the miss selection rate should not be more than 1%.
6 Test method
6.1 Overview
The test system is a stand-alone host, and the operating system is a Windows server series version, which can complete video playback and test results
Store. The video facial image extraction application system should provide configuration items for obtaining the URL of the video stream.
Each test item should be entered in the entire test video. During the test, the same video keeps the parameter settings unchanged, and different videos can be changed
Parameter test.
The function test item is tested twice in the same video, and the test result of each test should meet the requirements of 5.1, the average value of each test result of the performance test item
All should meet the requirements of 5.2.
6.2 Function test
6.2.1 Face detection
6.2.1.1 The face detection area setting function is tested as follows.
a) Turn off face clustering and face selection;
b) Use the F1 group standard video of A.2.2 in Appendix A, and enter the first paragraph F11;
c) Set the area where people appear in the screen as the detection area;
d) Play the standard video to complete the facial image extraction;
e) Refer to the B.1 output catalog in Appendix B to check whether the result is a human face image;
f) Set the area where no people appear on the screen as the detection area;
g) Replay the standard video to complete the facial image extraction;
h) Refer to the B.1 output catalog in Appendix B to check whether the result is a human face image;
i) Enter the next video in group F1;
j) Repeat c)~j) until the F11~F15 segment of video is input in sequence.
Determine whether the result meets 5.1.1a).
6.2.1.2 The face detection function of multiple persons is tested as follows.
a) Turn off face clustering and face selection;
b) Use the F2 group standard video of A.2.3 in Appendix A, and enter the first paragraph F21;
c) Set the area where multiple people appear in the screen as the detection area;
d) Play the standard video to complete the facial image extraction;
e) Refer to the B.1 output catalog in Appendix B to check whether there are face images of the same number of people in the detection area in the result;
f) Change the detection area where multiple people appear in the screen;
g) Replay the standard video to complete the facial image extraction;
h) Refer to the B.1 output catalog in Appendix B to check whether there are face images of the same number of people in the newly established area;
i) Enter the next video in group F2;
j) Repeat c)~j) until the F21~F25 segment of video is input in sequence.
Determine whether the result meets 5.1.1b).
6.2.1.3 The face size detection range setting function is tested as follows.
a) Turn off face clustering and face selection;
b) Set the full screen as the detection area;
c) Use the F1 group standard video of A.2.2 in Appendix A, and enter the first paragraph F11;
d) Do not set the face size detection range;
e) Play the standard video to complete the facial image extraction;
f) Refer to the B.1 output catalog in Appendix B to see if there are people with a distance between the eyes of no less than 100 pixels and no more than 10 pixels.
Face image;
g) Set the detection range of face size to the distance between eyes (30~60) pixels;
h) Replay the standard video to complete the facial image extraction;
i) Refer to the B.1 output catalog in Appendix B to see if the result is only in the range of not less than 30 pixels to not more than 60 pixels.
Face image within the enclosure;
j) Enter the next video in group F1;
k) Repeat d)~k) until the F11~F15 segment of video is input in sequence.
Determine whether the result meets 5.1.1c).
6.2.2 Face clustering
The face clustering on and off functions are tested as follows.
a) Turn off face selection;
b) Set the full screen as the detection area, without setting the face size detection range;
c) Use the F2 group standard video of A.2.3 in Appendix A, and enter the first paragraph F21;
d) Turn on face clustering;
e) Play the standard video to complete the facial image extraction;
f) Refer to the B.1 output catalog in Appendix B to check whether the facial images are grouped and identified according to the number of people entering the screen;
g) Turn off face clustering;
h) Replay the standard video to complete the facial image extraction;
i) Refer to the B.1 output catalog in Appendix B to check whether the facial images are not grouped and identified according to the number of people entering the screen;
j) Enter the next video in group F2;
k) Repeat d)~k) until the F21~F25 segment of video is input in sequence.
Determine whether the result meets 5.1.2.
6.2.3 Face selection
6.2.3.1 The face selection on and off function is tested according to the following methods.
a) Turn on face clustering;
b) Set the full screen as the detection area, without setting the face size detection range;
c) Use the F2 group standard video of A.2.3 in Appendix A, and enter the first paragraph F21;
d) Set predetermined requirements for selecting facial images;
e) Open face selection;
f) Play the standard video to complete the facial image extraction;
g) Refer to the B.1 output catalog in Appendix B to check whether the results only have face images that meet the predetermined requirements;
h) Close face selection;
i) Replay the standard video to complete the facial image extraction;
j) Refer to the B.1 output catalog in Appendix B to check whether there are face images that do not meet the predetermined requirements in the result;
k) Enter the next video in group F2;
l) Repeat d)~l) until the F21~F25 segment of video is input in sequence.
Determine whether the result meets 5.1.3a).
6.2.3.2 The function of setting the upper limit of the number of face images selected per person is tested as follows.
a) Turn on face clustering and face selection;
b) Set the full screen as the detection area, without setting the face size detection range;
c) Use the F2 group standard video of A.2.3 in Appendix A, and enter the first paragraph F21;
d) Set the upper limit of the number of face images selected per person to 1;
e) Play the standard video to complete the facial image extraction;
f) Refer to the B.1 output catalog in Appendix B to check whether only one face image is selected per person per time;
g) Set the upper limit of the number of face images selected per person to 5;
h) Replay the standard video to complete the facial image extraction;
i) Refer to the B.1 output catalog in Appendix B to see if only 5 face images are selected per person per time;
j) Enter the next video in group F2;
k) Repeat d)~k) until the F21~F25 segment of video is input in sequence.
Determine whether the result meets 5.1.3b).
6.2.3.3 Mandatory selection of the face image function that does not meet the requirements is tested according to the following methods.
a) Turn on face clustering and face selection;
b) Set the full screen as the detection area, and set the face size detection range to 80 pixels between the eyes;
c) Use the F3 group standard video of A.2.4 in Appendix A, and enter the first paragraph F31;
d) Enable forced selection of face images;
e) Play the standard video to complete the facial image extraction;
f) Refer to the B.1 output catalog in Appendix B to check if there are any facial images that do not meet the requirements;
g) Turn off forced selection of face images;
h) Replay the standard video to complete the facial image extraction;
i) Refer to the B.1 output catalog in Appendix B to check whether there is no unqualified face image in the result;
j) Enter the next video in group F3;
k) Repeat d)~k) until the F31~F35 segment of video is input in sequence.
Determine whether the result meets 5.1.3c).
6.3 Performance test
6.3.1 Face detection
When the false detection rate is not more than 1%, the performance of the missed detection rate is tested as follows.
a) Turn off face clustering and face selection;
b) Set the full screen as the detection area, without setting the face size detection range;
c) Use the P1 group standard video of A.3.2 in Appendix A, and enter the first paragraph P11;
d) Play the standard video to complete the facial image extraction;
e) Refer to the B.1 output catalog in Appendix B to view the results, and count the number of non-face images Bf and the detected images according to A.3.1.4 in Appendix A
Like the total number B;
f) Calculate Rfd according to formula (1);
g) Debug the parameters, enter the next video in group P1, repeat d)~g) input P11~P15 videos in sequence until the false detection rate is determined
Rfd value is not more than 1%;
h) Enter the first paragraph P21 of the P2 group standard video in A.3.3 in Appendix A;
i) Play the standard video to complete the facial image extraction;
j) Refer to the B.1 output catalog in Appendix B to view the results, and count people who have not detected face images according to A.3.1.4 in Appendix A
The number of Am and the total number of people that should be detected A;
k) Calculate Rmd according to formula (2);
l) Enter the next video in group P2;
m) Repeat i)~m) until the P21~P25 segment of video is completely input in sequence;
n) Calculate the average value of 5 missed detection rates Rmd.
Determine whether the result meets 5.2.1.
6.3.2 Face clustering
When the mispolymerization rate is not more than 5%, the performance of the leakage rate is tested as follows.
a) Turn off face selection and turn on face clustering;
b) Set the full screen as the detection area, without setting the face size detection range;
c) Use the P1 group standard video of A.3.2 in Appendix A, and enter the first paragraph P11;
d) Play the standard video to complete the facial image extraction;
e) Refer to the output catalog of B.1 in Appendix B to view the results, and according to A.3.1.4 in Appendix A, the clusters of face images of different people are calculated as
The number of groups in the same group Gf and the total number of cluster groups G;
f) Calculate Rft according to formula (3);
g) Debug the parameters, enter the next video in group P1, repeat d)~g) input P11~P15 videos in sequence until the misconvergence rate is determined
Rft value is not more than 5%;
h) Enter the first paragraph P21 of the P2 group standard video in A.3.3 in Appendix A;
i) Play the standard video to complete the facial image extraction;
j) Refer to the B.1 output catalog in Appendix B to view the results. According to A.3.1.4 in Appendix A, count the same-person face images that are not clustered as
The number of people in the same group Af and the total number of people who should be clustered A;
k) Calculate Rmt according to formula (4);
l) Enter the next video in group P2;
m) Repeat i)~m) until the P21~P25 segment of video is completely input in sequence;
n) Calculate the average value of Rmt for 5 times.
Determine whether the result meets 5.2.2.
6.3.3 Face selection
When the false selection rate is not more than 5%, the performance of the miss selection rate is tested as follows.
a) Turn on face clustering and face selection, and turn off forced selection;
b) Set the full screen as the detection area, without setting the face size detection range;
c) Use the P1 group standard video of A.3.2 in Appendix A, and enter the first paragraph P11;
d) Set the upper limit of the number of face images selected per person to 18;
e) Play the standard video to complete the facial image extraction;
f) Refer to the B.1 output catalog in Appendix B to view the results, and count the face images that do not meet the predetermined requirements according to A.3.1.4 in Appendix A
The number of images Cf and the total number of selected face images C;
g) Calculate Rfs according to formula (5);
h) Debug the parameters, enter the next video in group P1, set the upper limit of the number of face images selected per person per time to decrease 3 from the original value, repeat e)~
h) Input P11~P15 videos in sequence until it is determined that the value of the false selection rate Rfs is not greater than 5%;
i) Enter the first paragraph P21 of the P2 group standard video in A.3.3 in Appendix A;
j) Set the upper limit of the number of face images selected per person to 4;
k) Play the standard video to complete the facial image extraction;
l) Refer to the output catalog of B.1 in Appendix B to check the results, and according to A.3.1.4 in Appendix A, no face meeting the predetermined requirements is selected
The number of people in the image Dm and the total number of people who should be selected D;
m) Calculate Rms according to formula (6);
n) Set the upper limit of the number of face images selected per person to increase by 5 from the original value;
o) Enter the next video in group P2;
p) Repeat k)~p) until the P21~P25 segment of video is input in sequence;
q) Calculate the average value of 5 missed selection rates Rms
Determine whether the result meets 5.2.3.
6.4 Test file
See Appendix B for the test file record format.
The following content should be explained in the test report.
a) The test result or average value of each segment of the standard video;
b) Standard video format and the number of tested persons;
c) The behavior of the tested person in the standard video. single or multiple people, free walking or waiting statically, turning or bowing, multiple people crossing
Or overlap;
d) Recording scenes, indoor and outdoor, light intensity, backlight metering, facial light changes;
e) Take a screenshot of the recorded scene video.
Appendix A
(Normative appendix)
Video facial image extraction test standard video requirements
A.1 Video format, composition and numbering
A.1.1 Standard video should be recorded in H.264, SVAC or MPEG4 file format, recording 25 frames per second.
A.1.2 The definition of standard video should be 1080P or 720P.
A.1.3 Standard videos are divided into functional test videos and performance test videos. The functional test videos contain 3 groups, each group contains 5 segments, each group
They are used to test different function items; the performance test video contains 2 groups, each group contains 5 segments, one group is used to test the debugging parameters, and the other group is used to
Test different performance indicators.
A.1.4 Standard videos are represented by numbers, which are composed of class numbers, group numbers, and segment numbers. The function test video class starts with the letter F, and the performance test
The frequency category starts with the letter P, the category number and group number indicate the purpose of the video, and the segment number indicates the sequence of the video.
A.2 Function test video
A.2.1 Acquisition conditions
A.2.1.1 Indoor environment recording, the light is uniform and stable around the subject, the background is soft, there is no direct light and reflective spots, and the detection area
The illuminance is not less than.200lux.
A.2.1.2 The tested person keeps the front face, the horizontal rotation angle of the face does not exceed ±30°, the pitch angle does not exceed ±15°, and the tilt angle does not exceed
±10°.
A.2.1.3 The tested person has a neutral expression, no makeup, no face covering, and the bun does not cover the eyebrows.
A.2.1.4 The person being tested does not move laterally to avoid cross occlusion.
A.2.1.5 The tested person does not appear repeatedly in the same group of videos.
A.2.1.6 There is no other person in the scene except the person being tested.
A.2.2 Group F1 video
F1 group video recorded 5 segments, numbered according to F11~F15, used to test function items 5.1.1a) and 5.1.1c), each segment of content is continuously recorded
Less than 3min, and the total length is no less than 15min.
A tested person moves at a constant speed from far to near, and the distance between the eyes of his face image ranges from less than 10 pixels to greater than 100 pixels.
Continuous change.
A.2.3 Group F2 video
F2 group video recording 5 segments, numbered according to F21~F25, used to test function items 5.1.1b), 5.1.2, 5.1.3a) and 5.1.3b), each segment
The content is continuously recorded for no less than 3 minutes, and the total length is no less than 15 minutes.
The 15 subjects under test are arranged in five rows and three columns, with a distance of 5m from each other, moving at a constant speed from far to near.
10 pixels, the highest is greater than 100 pixels.
A.2.4 F3 group video
F3 group video recording 5 segments, numbered according to F31~F35, used to test function item 5.1.3c), each segment of content is continuously recorded for no less than 3 minutes,
The total length is not less than 15min.
A tested person moves at a constant speed from far to near, and the distance between the eyes of the face image ranges from a minimum of less than 10 pixels to a maximum of only 60 pixels.
Surrounding changes continuously.
A.3 Performance test video
A.3.1 Acquisition conditions
A.3.1.1 In the practical field collection indoors and outdoors, under the changing conditions of front light, side light or back light, the human face may have partial shadows
Or highlight changes.
A.3.1.2 It can be recorded in aisles, escalators, security gates, counters, elevator doors, sidewalks, etc.
A.3.1.3 Allow partial occlusion or partial occlusion of part of the face.
A.3.1.4 The cumulative number of people appearing in the video is no less than 400 people, and multiple people can appear in the scene at the same time.
For the tested person.
---The bun does not cover the eyebrows and does not wear a mask;
---At least there should be a face image with a distance between the eyes of no less than 30 pixels;
---At least, there should be images where the horizontal rotation angle of the human face does not exceed ±30°, the pitch angle does not exceed ±15°, and the tilt angle does not exceed ±10°.
All personnel that appear cumulatively are included in the compulsory extraction personnel, and the tested personnel are included in the extraction personnel.
A.3.2 P1 group video
P1 group video record 5 segments, numbered according to P11~P15, used to test performance items 5.2.1, 5.2.2 and parameter debugging before 5.2.3, each segment
The content is continuously recorded for no less than 3 minutes, and the total length is no less than 15 minutes.
A.3.3 P2 group video
The P2 group video recorded 5 segments, numbered according to P21~P25, used to test performance items 5.2.1, 5.2.2 and 5.2.3, each segment of content is continuously recorded
Less than 10min, and the total length is not less than 50min.
Tips & Frequently Asked Questions:Question 1: How long will the true-PDF of GA/T 1344-2016_English be delivered?Answer: Upon your order, we will start to translate GA/T 1344-2016_English as soon as possible, and keep you informed of the progress. The lead time is typically 1 ~ 3 working days. The lengthier the document the longer the lead time. Question 2: Can I share the purchased PDF of GA/T 1344-2016_English with my colleagues?Answer: Yes. The purchased PDF of GA/T 1344-2016_English will be deemed to be sold to your employer/organization who actually pays for it, including your colleagues and your employer's intranet. Question 3: Does the price include tax/VAT?Answer: Yes. Our tax invoice, downloaded/delivered in 9 seconds, includes all tax/VAT and complies with 100+ countries' tax regulations (tax exempted in 100+ countries) -- See Avoidance of Double Taxation Agreements (DTAs): List of DTAs signed between Singapore and 100+ countriesQuestion 4: Do you accept my currency other than USD?Answer: Yes. If you need your currency to be printed on the invoice, please write an email to [email protected]. In 2 working-hours, we will create a special link for you to pay in any currencies. Otherwise, follow the normal steps: Add to Cart -- Checkout -- Select your currency to pay.
|