A system and method for witness verification based on deep learning

Abstract

The present invention relates to a witness verification system and method, especially a witness verification system and method based on deep learning, which belongs to the field of security protection. The witness verification system based on deep learning includes a training subsystem and a witness verification sub-system system, the method of witness verification based on deep learning includes a training process and a witness verification process, the training process uses several known identities on-site face images and ID photos to train the training subsystem, and the witness verification The process uses the module parameters finally obtained in the training process to compare multiple on-site collected photos of passers-by with the certificate photos corresponding to the information collected by the certificate reading module, and automatically output the verification results of whether the personnel and certificates match. The certificate photo is compared with multiple face photos collected by on-site video surveillance equipment. It has strong adaptability to factors such as face angle, on-site environment, and lighting, and has high accuracy of witness verification and fast passing speed.

Description

technical field [0001] The invention relates to a witness verification system and method, in particular to a deep learning-based witness verification system and method, belonging to the security field. Background technique [0002] In places such as railway stations, bus stations, public security checkpoints, and large-scale exhibitions, due to security or other business needs, it is often necessary to check the credentials of passers-by on the spot. At present, this kind of verification work is mainly done manually by the staff, and it is impossible to prevent criminals from using other people's ID cards to muddle through the customs, and there are potential safety hazards. [0003] In this situation, a verification method based on face recognition technology has emerged. For example, the patent document with application number 201310099064.7 discloses an authentication system and method based on face recognition for the second-generation resident ID card, including The se...

AI Technical Summary

Problems solved by technology

This type of method and system has the following disadvantages: it usually only compares one photo collected with the ID photo, cannot effectively use the information of multiple face photos collected from different angles when people pass, and sometimes requires people to stop and cooperate Taking pictures reduces the traffic speed; and this system has low adaptability to factors such as ambient light and face angles, and the accuracy rate is not high in practical applications; in addition, this method cannot be specially optimized for the ID photo database. The similarity between on-site monitoring photos and ID photos is relatively low; existing face recognition systems often rely on the precise positioning of facial contour feature points, but in practical applications, facial occlusion, lighting changes and other reasons will lead to feature points The positioning is inaccurate, or the generated facial network is inaccurate, resulting in a decline in subsequent recognition accuracy

Images