GA/T 1154.4-2018Video analysis instrument - Part 4: Technical requirements for face analysis (English PDF)

1 Scope 1

2 Normative references 1

3 Terms, definitions and abbreviations 1

4 Product composition description 3

4.1 Product code and number 3

4.2 Composition block diagram 3

4.3 Reference Logical Relationship Model 3

5 Technical requirements 4

5.1 General requirements 4

5.1.1 General requirements 4

5.1.2 Software interface requirements 4

5.2 Functional requirements 4

5.2.1 Face video image input/import 4

5.2.2 Face database 5

5.2.3 Face analysis 5

5.2.4 Output of face analysis results 7

5.2.5 Interfacing with public security video image information application system 7

5.3 Performance requirements 7

5.3.1 Face database 7

5.3.2 Face detection 7

5.3.3 Face comparison 7

5.3.4 Face Retrieval 8

5.3.5 Face Control 8

6 Test method 8

6.1 Test conditions 8

6.2 Functional requirements inspection 8

6.2.1 Face video image input/import inspection 8

6.2.2 Face database verification 8

6.2.3 Face analysis test 9

6.2.4 Output verification of face analysis results 10

6.2.5 Docking inspection with public security video image information application system 10

6.3 Performance requirements inspection 10

6.3.1 Data preparation 10

6.3.2 Face database verification 11

6.3.3 Face detection inspection 11

6.3.4 Face comparison test 11

6.3.5 Face retrieval test 12

6.3.6 Face control inspection 12

7 Inspection rules 12

Appendix A (Normative Appendix) Rest Interface Specification 14

Appendix B (informative appendix) Video face description 16

Appendix C (informative appendix) Face video image test sample 20

Reference 22

Video image analyzer Part 4: Technical Requirements for Face Analysis

1 Scope

This part of GA/T 1154 specifies the technical requirements, test methods and inspection rules for face analysis of video image analyzers.

This section applies to the development, production and inspection of video image analyzers with video image face analysis functions.

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.

GB/T 31488-2015 Security and protection video surveillance face recognition system technical requirements GA/T 922.2-2011 Security Face Recognition Application System Part 2: Face Image Data GA/T 1154.1-2014 Video Image Analyzer Part 1: General Technical Requirements GA/T 1154.3-2017 Video image analyzer Part 3: Video image retrieval technical requirements GA/T 1325-2017 Video image collection specification for security and face recognition applications GA/T 1326-2017 Security Protection Face Recognition Application Program Interface Specification GA/T 1344-2016 Technical requirements for face image extraction from video of security face recognition applications GA/T 1400.3-2017 Public Security Video Image Information Application System Part 3: Database Technical Requirements GA/T 1400.4-2017 Public Security Video Image Information Application System Part 4: Interface Protocol Requirements

3 Terms, definitions and abbreviations

3.1 Terms and definitions GB/T 31488-2015, GA/T 1344-2016, GA/T 1154.1-2014 and the following terms and definitions apply to this document.

3.1.1

Face detection

The process of finding a human face from a given dynamic video or static image and determining its position and size.

3.1.2

Faceenrolment

The process of extracting and storing the characteristics of the input face image, and assigning its identity.

3.1.3

Faceattributeanalysis the attribute information and the additional feature information of the sample face extracted from the face image sample.

Note 1.Attribute information, such as. age, gender, etc.

Note 2.Additional feature information, such as whether to wear a hat, whether to have a beard, whether to wear glasses, whether to wear a mask, whether to wear accessories, etc.

3.1.4

Face recognition

The process of detecting faces from video images, extracting face features and comparing them.

3.1.5

Facesearching

Based on face recognition and comparison technology, enter the face image to be retrieved, and search for the same image from the registered face image library or video extraction image library.

A process of face image and output retrieval results.

3.1.6

Facealert

It is determined that there are people on the watch list in the video image, and a prompt message is output.

3.1.7

Response time

The time interval from when the face video image to be compared is input to the analyzer to the result output.

3.1.8

Enrolmentfailurerate the percentage of the number of failed registration images to the number of imported images.

EFR=

N ×100%

Where.

EFR --- registration failure rate;

B --- number of successes;

N ---The number of imported images.

3.1.9

Top N identification rate topN identification rate; rankN Correctly identify the percentage of the number of samples with the top N results to the total number of test samples.

Top N recognition rate = Correctly identify the number of samples in the top N the total number of test samples × 100% where.

N---positive natural number.

3.1.10

True positive

The target that should be retrieved appears in the video image, and the video image analyzer outputs the correct retrieval result for the target.

3.1.11

Falsenegative

The target that should be retrieved appears in the video image, but the target is not included in the retrieval result output by the video image analyzer.

3.1.12

False positive

The target that should be retrieved appears in the video image, but the video image analyzer outputs the wrong retrieval result for the target.

3.1.13

Detection rate truepositive(recal)rate the video image analyzer retrieves the correct target output and the video image should be retrieved by the video image analyzer.

Points ratio.

GA/T 1154.3-2017 Appendix A (30 video formats in Appendix A), the video can be played normally after transcoding, and there is no blur, black screen, Abnormal conditions such as screen jam.

5.2.2 Face database

5.2.2.1 General requirements should support the establishment of a registration database, a snapshot database and other face databases for different purposes.

5.2.2.2 Register library it should support the establishment of a face registration library to achieve the following functions.

a) Support the use of static face image data that meets the requirements of 4.1 in GA/T 922.2-2011 for face registration;

b) Support on-site collection of face image data for face registration;

c) Support batch import of face image data for face registration;

d) Support management operations such as adding, querying, modifying, and deleting registered libraries, and support sub-library management of multiple libraries;

e) Support the import and export of the registered library.

5.2.2.3 Snapshot library it should support the establishment of a face capture library to achieve the following functions.

a) Support the capture and storage of dynamic face video file data;

b) Support management operations such as adding, querying, modifying and deleting snapshot libraries, and supporting sub-library management of multiple libraries.

5.2.3 Face analysis

5.2.3.1 Face detection Face detection should meet the following requirements.

a) With video image selection function, selective image or video face detection;

b) It has the function of setting the face detection area, only detecting faces in the area;

c) With multi-face detection function, it can detect multiple faces in the detection area;

d) It has the function of setting the detection range of face size, and only detects faces that meet the requirements.

5.2.3.2 Analysis of face attributes the face attribute analysis should meet the following requirements.

a) With opening and closing functions;

b) Support the analysis of the attribute information of the face in the video image, including but not limited to age group, gender, etc.;

c) Support the analysis of the additional feature information presented by the face in the video image, including but not limited to whether to wear a hat, whether there is a Hu Child, whether to wear glasses, whether to wear a mask, whether to wear accessories, etc.;

d) Support the function of saving or outputting analysis results;

e) Video structured semantic information description should be generated, and meet the requirements of GA/T 1400.3-2017.

5.2.3.3 Facial statistical analysis should support the statistical analysis of the faces in the video according to the number, frequency, attributes, etc.

5.2.3.4 Face comparison the face ratio corresponds to the following requirements.

a) 1.1 comparison. compare two face images to generate similarity scores;

b) 1.N comparison. compare the input 1 face image with the N face images in the face database to generate similarity scores Value, sorted in descending order by similarity;

c) n. N comparison. compare the input n face images with the N face images in the face database to generate similarity scores Value, sorted in descending order by similarity.

5.2.3.5 Face Search

5.2.3.5.1 Image-based face retrieval Image face retrieval should meet the following requirements.

a) Support searching in the face database based on the target face image;

b) Support outputting images similar to the target face in order of similarity;

c) Support automatic face retrieval for a single target image containing multiple faces;

d) Support automatic batch face retrieval of multiple target images;

e) Support the export of image face retrieval results.

5.2.3.5.2 Video-based face retrieval Video face retrieval should meet the following requirements.

a) Support searching in the face database based on the target face video;

b) Support outputting video information similar to the target face in order of similarity;

Note. The video information includes the image information of the target face in the video (face image, panoramic image, image frame number), and the target face appears in the video.

The current time period.

c) Support export of video face retrieval results.

5.2.3.5.3 Face retrieval based on face attributes Face retrieval based on face attributes should meet the following requirements.

a) Support searching according to the attribute information of the face, including but not limited to age group, gender, etc.;

b) Support searching based on the additional feature information presented by the face, including but not limited to whether you wear a hat, whether you have a beard, whether Wear glasses, whether to wear a mask, whether to wear accessories, etc.;

c) Supports secondary face attribute retrieval of face retrieval results.

5.2.3.6 Face deployment control

5.2.3.6.1 Watchlist management Watchlist management should meet the following requirements.

a) It should have the management function of the watch list, and realize the management operations of adding, querying, modifying and deleting the watch list;

b) It should support the management function of multiple watch lists.

5.2.3.6.2 Alarm output the alarm output should meet the following requirements.

a) According to the attention target deployment control information, output the alarm scene image, the scene face image and the corresponding target face image in the attention list Images and related information;

b) It is advisable to output the similarity score between the live face image and the target face image in the watch list.

5.2.3.6.3 Alarm record management the alarm record should meet the following requirements.

a) Real-time recording of alarm captured images marked with the location of the target face, corresponding to the target face image and related information in the watch list, similar Degree points, alarm time. The alarm time should include. year, month, day, hour, minute, and second, and the year should be in millennium notation.

b) It has functions such as query, statistics and export of alarm records.

5.2.4 Output of face analysis results should support video image face analysis results refer to Appendix B format output description file.

5.2.5 Docking with public security video image information application system the connection with the public security video image information application system shall meet the following requirements.

a) Data exchange should be carried out with the public security video image information application platform through the analysis interface;

b) It is advisable to exchange data with the public security video image information database through the data service interface or collection interface, which should comply with Regulations in GA/T 1400.3-2017;

c) The analysis interface, data service interface and acquisition interface protocol should comply with the regulations in GA/T 1400.4-2017;

d) The face analyzer description data should support SDK (software development kit), and the data format adopts XML or Json.

Type package, see Appendix B for specific requirements.

5.3 Performance requirements

5.3.1 Face database

5.3.1.1 Face database capacity the face database capacity should meet the following requirements.

a) Registered library capacity. no less than 300,000 images;

b) The size of the snapshot library. no less than 1 million images.

5.3.1.2 Face image quality the quality of the face images stored in the database should meet the requirements of 4.3.2 in GA/T 1325-2017.

5.3.1.3 Face registration when meeting the requirements of 5.3.1.2, the face registration failure rate (EFR) should not be greater than 0.1%.

5.3.2 Face detection when the false detection rate is not more than 1%, the missed detection rate should not be more than 5%.

5.3.3 Face comparison 5.3.3.1 1.1 face verification when the false face acceptance rate is 0.1%, the false rejection rate does not exceed 5%.

5.3.3.2 1.N face recognition 1.N static face recognition, the top 10 recognition rate is not less than 90%.

5.3.4 Face Search

5.3.4.1 Image-based face retrieval in the snapshot library, the top 10 hit rate of image-based face retrieval is not less than 85%.

5.3.4.2 Video-based face retrieval in the snapshot library, the top 10 hit rate of video-based face retrieval is not less than 85%.

5.3.4.3 Face retrieval based on face attributes in the snapshot library, the average face retrieval accuracy (AP) based on face attributes is not less than 85%, and the following single face attribute retrieval accuracy rates at least not less than 75%.

a) Age group analysis;

b) Gender analysis;

c) Analysis of whether to wear a mask;

d) Whether to wear glasses for analysis;

e) Whether to wear a hat/accessories analysis.

5.3.5 Face deployment control when the false alarm rate of non-following lists is not higher than 5%, the under-reporting rate of follow-lists is not higher than 15%, and the accuracy of follow-list recognition is not less than 85%.

6 Test method

6.1 Test conditions

It shall meet the requirements of 6.1 in GA/T 1154.1-2014.

6.2 Functional requirements inspection

6.2.1 Face video image input/import inspection

6.2.1.1 Verification of face video image acquisition mode Collect separately according to the collection method of 5.2.1.1 and judge whether it meets the requirements.

6.2.1.2 Video image format verification Import a copy of the original video image samples of various video image formats specified in 5.2.1.2 to the face analyzer for decoding, and determine Whether it meets the requirements of 5.2.1.2.

According to Appendix A of GA/T 1154.3-2017, 30 original video image samples in different formats are provided and imported into the face analyzer for broadcasting Release and determine whether it meets the requirements of 5.2.1.2.

6.2.2 Face database verification the face database inspection should be carried out according to the following steps.

a) Add, query, modify, delete, import, export and other operations to the face registration database to determine whether it meets the requirements of 5.2.2.2;

b) Add, query, modify, delete, and sub-database operations on the face capture library to determine whether it meets the requirements of 5.2.2.3.

6.2.3 Face analysis test

6.2.3.1 Face detection inspection the inspection of the face detection function should be carried out in the following steps.

a) Set conditions such as face detection area, multiple faces, and face size detection range;

b) Select a video or image, perform face detection, and view the output result to determine whether it meets the requirements of 5.2.3.1.

6.2.3.2 Analysis and verification of face attributes the verification of the face attribute analysis function should be carried out in the following steps.

a) Turn on the face attribute analysis function;

b) Select the video, analyze the face attributes, and view the output result to determine whether it meets the requirements of 5.2.3.2.

6.2.3.3 Statistical analysis of human faces the test of face statistical analysis function should be carried out according to the following steps.

a) Turn on the face statistical analysis function;

b) Select the video, perform face statistical analysis, and view the output result to determine whether it meets the requirements of 5.2.3.3.

6.2.3.4 Face comparison test the verification of the face comparison function should be carried out in the following steps.

a) Choose a 1.1 comparison of two face images, check the similarity score, and judge whether it meets the requirements of 5.2.3.4a);

b) Choose to input 1 face image, and compare it with N face images in the face database to check the similarity score and the descending sort result, judge whether it meets the requirements of 5.2.3.4b);

c) Choose to input n face images, and compare them with N face images in the face database to check the similarity score and the descending sort result, judge whether it meets the requirements of 5.2.3.4c).

6.2.3.5 Face retrieval test

6.2.3.5.1 Image-based face retrieval test the inspection of the image-based face retrieval function should be carried out in the following steps.

a) Use the face registration function to register the face image of the test subject in the face database;

b) Select different face images of the test subjects to perform face retrieval in the face database;

c) View the search output result and judge whether it meets the requirements of 5.2.3.5.1.

6.2.3.5.2 Video-based face retrieval test the video-based face retrieval function test should be carried out in the following steps.

a) Use the face registration function to register the video containing the face of the test subject into the face database;

b) Select different face videos of the test subjects to perform face retrieval in the face database;

c) View the search output result and judge whether it meets the requirements of 5.2.3.5.2.

6.2.3.5.3 Face retrieval test based on face attributes the verification of the face attribute retrieval function should be carried out in the following steps.

a) Set the search conditions for face attributes;

b) Select images or videos from the face database as the search scope to search for face attributes;

c) View the search results, display methods and export result information, and judge whether it meets the requirements of 5.2.3.5.3.

6.2.3.6 Face control inspection

6.2.3.6.1 Watchlist management inspection the inspection of the watch list management function should be carried out according to the following steps.

a) Select the face image or video of the test subject, add it to the watch list, and maintain the relevant information of the target person;

b) Check the watch list and whether the target person can be modified, deleted, and searched for management;

c) View the management of multiple watch lists.

Determine whether it meets the requirements of 5.2.3.6.1.

6.2.3.6.2 Alarm output check Perform an analysis test on the registered test object, check whether the watch list is alarmed and related alarm information, and judge whether it meets 5.2.3.6.2 requirements.

6.2.3.6.3 Alarm record management inspection Perform an analysis test on the registered test object, view alarm information, and perform alarm record query, statistics, and export operations.

To determine whether it meets the requirements of 5.2.3.6.3.

6.2.4 Output verification of face analysis results Check the video image face description file to determine whether it meets the requirements of 5.2.4.

6.2.5 Docking inspection with public security video image information application system the inspection of the connection with the public security video image information application system should be carried out according to the following steps.

a) The face analyzer is connected to the public security video image information application platform, and data is exchanged through the analysis interface;

b) The face analyzer is connected to the public security video image information database, and data is exchanged through the data service interface or the acquisition interface;

c) Check whether the interaction protocol and description data meet the requirements of 5.2.5.

6.3 Performance requirements inspection

6.3.1 Data preparation

6.3.1.1 Data category calibration refer to the standards in Appendix C, Table C.1, calibrate the data complexity level of the test sample data source, and perform the calibration separately according to the calibration results Preparation of video image data.

6.3.1.2 Image data preparation Construct a face image test data set, each person has 1 face image, the male to female ratio is 1.1, and the age is 16 to 60 years old, accounting for 70%, less than 16 years old accounted for 15%, more than 60 years old accounted for 15%.

Construct a face image detection data set, with 1 face image per person, the face image set used for detection here is the number of face image tests belonging to the data set is different from the test face image set.

6.3.1.3 Video data preparation Construct a face video test data set, select video data for registration, and obtain at least 1 face image per person in the video, male to female ratio 1.1, 70% are 16 to 60 years old, 15% younger than 16 years old, 15% older than 60 years old.

Construct a face video detection data set, each person in the video obtains at least 1 face image, here the effective target set in the detection video is the official A video belonging to the face video test data set but different from the test sample set.

6.3.2 Face database verification

6.3.2.1 Capacity test of face database Import the face registration data set to determine whether the capacity of the face registration library meets the requirements of 5.3.1.1a).

Import video face test samples to determine whether the capacity of the face capture library meets the requirements of 5.3.1.1b).

6.3.2.2 Face image quality inspection in accordance with the method of GA/T 1325-2017 5.3.2, the inspection is carried out to determine whether it meets the requirements of 5.3.1.2.

6.3.2.3 Face registration inspection Register the 300,000 face registration data set into the system, record the number of successful imports, calculate the registration failure rate (EFR), and determine whether it meets 5.3.1.3 requirements.

6.3.3 Face detection inspection Import the video face test samples into the face analyzer, perform face detection and analysis, and count the number of positive detections, missed detections, and false detections of face detection Count, calculate the missed detection rate and false detection rate of face detection, and judge whether the result meets the requirements of 5.3.2.

6.3.4 Face comparison test 6.3.4.1 1.1 face verification test

6.3.4.1.1 Test data volume the number of face images in the face image library test set. 10,000.

The number of face images in the corresponding detection set. 10,000.

6.3.4.1.2 1.1 face verification comparison index statistics Take 10,000 test set samples as personal images, randomly select 10 from the 10,000 detection sets, and form 100,000 pairs of negative samples I compare it, output the comparison similarity results and sort them in descending order, and find the similarity at 0.1% false acceptance rate as the threshold.

Comparing 10,000 pairs of positive samples from the 10,000 test set and 10,000 detection set face images, count the 0.1% false acceptance rate threshol...