The Research And Implementation Of Face Recognition Based On MTCNN Framework

With the rapid development of artificial intelligence,vehicle-mounted face detection and recognition technology is also constantly updated and developed.In the vehicle-mounted face recognition system of the vehicle,we should take into account the limitations of the vehicle environment for the vehicle-mounted face recognition system,and also take into account the detection accuracy and recognition speed of the vehicle-mounted face recognition system as well as the running speed on the hardware.Nowadays,face recognition technology is susceptible to the slow running speed of the machine,the Angle and light of the acquired face pictures are affected by these problems,which may lead to the non-representative feature point threshold of face feature extraction or the failure of face detection system to detect faces.The reasons for the above problems are many,such as the lack of representativeness of training samples,the lack of fine image processing,the relatively simple feature point extraction algorithm and comparison,and the limitation of its hardware equipment.Therefore,it is particularly important to enhance the accuracy and robustness of feature points in face detection.In this paper,the speed of face recognition technology in hardware and its own face detection and recognition algorithm are studied.The main work contents are as follows:(1)To make the early samples reach a balanced state.In the partitioning process,the data set is expanded for the samples,and the third party's model is used to participate in the detection and labeling of the data set.Then,the data set is used for training to generate a convertible operator model.(2)In the process of face matching and comparison,the conventional MTCNN +Facenet algorithm is used to integrate face recognition on the hardware,but the detection and comparison time is long,and the speed of the algorithm is limited by the hardware configuration;Then use the hash comparison algorithm to find that its algorithm has a higher requirement for the environment,finally use the way of Mt CNN call FACE++ for face recognition,and successfully achieve face comparison and feature extraction,according to the data results returned by face recognition in three ways,it is found that MTCNN call FACE++ has a lower requirement for configuration.At the same time,its detection speed and accuracy are the best among the three algorithms.(3)Based on the basic use of face recognition in the way of MTCNN call FACE++,the training framework of MTCNN is improved,and the frame limit of MTCNN call FACE++ is adjusted.Finally,the relatively high speed face recognition is well realized in hardware.