Design And Implementation Of Vehicle Identity Authentication System Based On Face Recognition

Nowadays,face recognition is very popular in the field of identity authentication and it has proved its significance in practice.In this thesis,an economical,safe and robust vehicle-mounted face recognition system is implemented.The system firstly performs face detection and face liveness detection on the face image.After passing the face liveness detection,the identity of the commercial vehicle driver is verified through face recognition.The main work of this thesis is as follows:(1)In the face detection task,Center Face is simplified,the multiple heads of Center Face are integrated into one head and the bottleneck module in the Mobile Net backbone network structure uses two depthwise convolutions,besides,a multi-scale training method is introduced to reduce the size of the model while ensuring the accuracy of the model;A near-infrared camera is used to improve the robustness of face detection;By introducing maximum face detection and face detection area of interest,it ensures that the driver is detected in the multiple faces that may appear in the cockpit.(2)In the task of face liveness detection,the imaging characteristics of near-infrared cameras,the HSV characteristics of visible light images and the HOG characteristics of near-infrared images are used to achieve an efficient face liveness detection method to ensure the efficiency of the system.(3)In the face recognition task,a near-infrared camera is used to improve the robustness of the system’s recognition and the Mobile Face Net network of the Insightface model is optimized,the bottleneck module also uses two depthwise convolutions and the best loss function and parameters are selected through experimental tests to make the model perform optimally.(4)Based on the above technical research,an identity verification system based on face information is designed and implemented,which can realize automatic identity verification based on binocular cameras and the final experimental results show that the solution in this thesis is more efficient and can meet the established requirements of this thesis.