Overview of the background of the integration of human and evidence
With the acceleration of urbanization in China, urban population is becoming more and more intensive, and the mobility of population is also greatly increased. The social crime rate is increasing year by year. In the traditional way of investigation, manual investigation is often used. It is not only time-consuming and laborious to check the identity of personnel in important places, but also to restrict the entry of foreign personnel into fixed areas, which may cause omissions and other situations. The efficiency of investigation is greatly reduced, and at the same time, it brings great difficulties to public security prevention and social stability maintenance.
In order to effectively solve the social governance problems in key complex areas, consolidate the basic foundation of social stability and long-term stability, and the growing popularity and maturity of high-definition technology, intelligent technology and network technology, our company, based on actual needs, launched a face recognition system solution for the situation of large floating population, difficult identification verification and difficult management of personnel visits in complex areas.
The system uses advanced face recognition algorithm and high-speed chip as the operation hardware platform of recognition algorithm. Through the access of ID card information collection, real-time face capture and human card comparison, the human card integration verification is realized. And for different places to achieve fixed personnel brush face pass, visitor personnel card comparison registration, to solve the problem of fixed personnel every time need to brush or input password, person card comparison failure needs security personnel or staff manual confirmation and release.
System design principles
The design of the application scheme follows the principle of advanced technology, multiple algorithms, stable performance and cost saving; the design content of the system is systematic, complete and comprehensive; the design scheme is scientific, reasonable and operable.
It is based on the core algorithm of the following authentication system:
High speed face recognition algorithm
People who have the common sense of sketch know that the area of golden pentagons formed from eyebrow arch to cheekbone to nasal bone and upper jawbone does not move. No matter what exaggerated expression is made, no matter how fat or thin changes are, no matter what angle, this area maintains a relatively fixed geometric relationship. Based on these invariable factors, we use the method of deep learning to invent the bone point pentagonal face recognition algorithm, which is free from the interference of skin color and race. At the same time, it can well avoid the influence of expression change, beard camouflage, angle deviation, hair change, glasses decoration, mouth occlusion and other environmental factors on face recognition.
G represents our basic idea of algorithm "Gu", and also represents the meaning of "global" and "great" all over the world
Age and gender algorithm -- the difficulty of ID card recognition
With the increase of age, people's face structure will change slightly. The face shape in childhood is different from that in adolescence, and it is different in youth and after middle age. But the biggest changes are from children to youth, from middle age to old age. For example, if the cheek grows longer, the mandible, eyebrow bone, parotid flesh, maxilla and lip will have relatively large changes, plus wrinkles, hair or scars, etc., it will change the face, and then affect the effect of face recognition. The validity period of ID card is generally 5-20 years. How to solve the problem of age span is the fatal factor of face recognition effect.
From the perspective of easier age estimation and gender judgment, Dongwo intelligent algorithm team developed a new recognition algorithm agface, which significantly improved the effect of cross age face recognition. Specifically, this new algorithm uses a competitive task learning algorithm to combine the two tasks in a common optimization framework. In this way, the information obtained from age estimation can effectively help the feature selection of cross age face recognition
Combination of multiple algorithms -- a new direction of face recognition
Combined with several face recognition algorithms, as well as age algorithm, gender algorithm, color algorithm, tracking algorithm and other recognition algorithms, it can effectively improve the accuracy of recognition, reduce false alarm interference, and make dynamic face recognition meet the actual use standard.
Face recognition algorithm standard of ID card
License plate recognition algorithm
Combined with several license plate recognition algorithms, as well as character algorithm, color algorithm, type algorithm and other recognition algorithms, it can effectively improve recognition accuracy and reduce false alarm interference.
System composition diagram
Face bayonet system
The system is composed of big data platform, verification terminal, on-site client and information integrated platform. The blacklist data is sent to the self-service verification terminal under the information integrated platform, and the face photos are taken and extracted by the personal certificate verification terminal, and uploaded to the big data collection platform. The on-site law enforcement client checks the customs clearance, and makes the suspect warning processing
License plate person card integration system
The system consists of data acquisition server, management platform, key personnel data server, high-end automatic acquisition terminal, network camera, on-site client and remote user. Key population data servers send blacklist data to high-end automatic collection terminals, high-end automatic collection terminals take photos to extract face photos and license plate information, upload them to big data collection platform, on-site law enforcement clients check customs clearance, and do suspect early warning processing
System functions and advantages
High accuracy, working offline
The world's leading algorithm completely solves the problems of face cross age, face small picture recognition, multiple types of license plate recognition, and face does not need to connect to the public security network to access the large picture of the certificate, so as to identify the authenticity of the certificate and whether the person is the same as the vehicle