Face recognition technology system "Skynet Engineering" application solution

The face recognition function can be divided into two aspects according to the source of the face data: the static face comparison function (face large library search) and the dynamic face comparison function (personal bayonet black list real-time comparison).

First, the demand

The "Skynet Project" face recognition system includes a series of face recognition technology-based application systems such as a face capture comparison system, a face retrieval system, and a post-video retrieval system (human face). These systems mainly use face detection algorithms, face tracking algorithms, face quality scoring algorithms and face recognition algorithms; specifically for personnel monitoring, troubleshooting and retrieval, are video analysis, motion tracking, face detection and recognition technology in the security field New integrated application.

The face large library search function refers to importing the face information database such as the urban resident population database, the suspect escape database or the face capture database into the face recognition server, and the server models and analyzes the face image. When you input a face photo to the system, you can search for the identity of the person who is similar to the input photo in the imported database. At the same time, the system can also check the duplicate personnel in the single face database or the multi-face database that have been imported into the database and generate the troubleshooting results.

The real-time comparison function of the personnel bayonet refers to the face capture of the person passing through the bayonet by deploying and installing a camera card at a key monitoring point in the city. The front camera captures the captured face image through the computer network to the database of the monitoring center for data storage, and compares it with the face blacklist library in real time. When a suspicious person is found, the system automatically sends out an alarm signal, using various linkages. The way to notify the staff on duty. The system integrates high-definition face image capture, transmission, storage, face feature extraction and analysis and recognition, automatic alarm and network control and other functions in one, and has powerful query function and powerful communication and networking functions, which can be widely used. Pedestrian monitoring at important levels, together with the vehicle bayonet system, can be used to construct a modern intelligent monitoring system with vehicle and pedestrian monitoring.

Second, the system architecture

The face recognition system mainly adopts face detection algorithm, face tracking algorithm, face quality scoring algorithm and face recognition algorithm. The utility model realizes the functions of capturing and capturing, modeling and storing, real-time blacklist comparison alarm and face large library retrieval of the face of the entering and leaving personnel.

Third, the system consists of

The face information integrated application platform is composed of a face dynamic comparison unit, a human trajectory analysis unit, a personnel identification unit and a video storage unit.

* Face dynamic comparison unit

The face dynamic comparison unit is deployed in the video private network, mainly based on the face modeling algorithm with independent intellectual property rights and the face comparison algorithm, which can be connected to the front face face capture machine to provide a real-time blacklist comparison alarm for the face. The function realizes the early deployment and real-time warning of terrorism, stability, and criminals, and grasps the dynamics of Zhian in real time.

* Personnel trajectory analysis unit

The personnel trajectory analysis unit is deployed in the video private network, and can be connected to the front-end face capture machine to model and store the captured data returned by the front-end face capture machine, and establish a massive facial feature database; and support the capture of the face. The library's fast retrieval and graph search function, using the unique YI feature of face recognition features, accurately and efficiently realize the effective investigation of the massive face photo database, helping the public an min to quickly lock the suspect's activity track and improve Case investigation efficiency.

* Personnel identification unit

The personnel identification unit is recommended to be deployed in the Gongan Information Network. Through the docking with the gongan business library, the static face feature database can be established by importing the resident population database, the temporary resident population photos and personnel identification information, and supporting the static person. The quick retrieval of the face database and the search of the map can realize the quick confirmation of the identity of the unknown person. At the same time, it supports the mobile terminal such as the shou machine and the police service of the Android system to perform remote face retrieval and quickly confirm the identity of the person.

* Video storage unit

The real-time monitoring video collected by the front-end face capture machine can be written into the video storage unit through the streaming media direct write mode. The streaming media direct storage technology can improve system performance and reliability, and has the characteristics of high density, large capacity, and easy expansion. The video storage unit supports multiple protocols such as RTSP, GB/T28181, ONVIF, and PSIA. It integrates encoding device management, video recording management, storage and forwarding functions. The platform and client can directly download and download from the storage unit.

Fourth, system function

* Face dynamic comparison warning

Supporting the introduction of portraits and information of local drug users, fugitives and former staff members, as well as other local control personnel, including name, gender, fen certificate number, home address, face photo, etc. The above-mentioned personnel's early control (can be controlled according to time, location, distribution control level, acquaintance alarm threshold and other information). The system can compare the face captured by the front-end face capture machine with the cloth control library in real time. When the similarity between the face of the captured face and the control library reaches the set alarm threshold, the system automatically alarms in real time.

* Personnel trajectory analysis tracking

1, feature search

The system can model and store the captured data returned by the front-end face capture machine, build a massive facial feature database, and support the characteristics of the gender, age, and wearing glasses, as well as the time and place of the capture. Capture people for a quick search.

2, photo search

Enter a face photo (document photo or clear face image captured by Zhian monitoring), in the massive face capture library, according to the portrait feature point comparison algorithm, retrieve the most similar person, according to the similarity The high to low are arranged in order, and the location and time information of the captured are displayed, thereby helping the police to quickly lock the suspect's activity track.

3, frequent analysis

According to the capture time period and the position of the capture card point, in the massive face capture library, the collision comparison is performed according to the similarity condition, and the similar face appearing at the time period and position is found, thereby analyzing the abnormal person of the activity, and discovering at the same time Suspects will frequently appear in other characteristics before they step on the scene.

4. Regional collision analysis

For the specified two or more area ranges, in the massive face capture library, the collision comparison is performed according to the similarity condition, and the similar face appearing in the above-mentioned multiple areas is captured in the set time. In order to quickly confirm the locking of suspicious personnel, provide key clues for serial analysis, such as serial cases and case types.

* Personnel identification query

1, identification identification

Enter a face photo (a face photo captured by the front-end face grabber or a clear face image captured by Zhian), which can be found in the static face feature database (by importing the resident population database, temporary resident population photos) And the establishment of the identity information of the person), according to the portrait feature point comparison algorithm, the most similar personnel are retrieved, and the similarity is ranked from high to low, which helps the public to quickly lock the identity information of the unknown person.

At the same time, it supports Android mobile phone terminals such as shou machine and police service to perform remote face retrieval. It supports on-site capture of photos and automatic matching of photos in the database to quickly confirm the identity of the person.

2, identity comparison check

The system supports N:N comparison search for personnel information such as permanent resident population, temporary resident population database, and body census library, and finds out the list of personnel who repeatedly apply for fen certificate, household registration, temporary residence permit, etc.

At the same time, the system can help to quickly confirm the possibility of the suspect involved in other cases by comparing the face photos of the suspects in the ordinary case with the fugitives in the escaping library; or the N:N way of carrying out personnel information for the national escaping personnel library. The comparison search provides relevant clues for the actions of the Gongan authorities in the escape network.

The results of the check are graphically displayed to facilitate manual verification of the results of the investigation.

V. Conclusion

Face recognition technology has developed to this day, and there are still some limitations. Light, posture, expression, age, etc. have a greater impact on face comparison. Face recognition has a high demand for scenes and can only be used in specific scenes. With the development of 3D face scanning technology and the continuous improvement of deep learning algorithms, future face recognition will be more widely used in various fields such as finance and transportation, popularizing people's daily lives, and realizing the security precautions. The role of bringing security to people's work and life.