A Face Recognition Conference Attendance System Based on Deep Learning

Face recognition based on deep learning is an important application of artificial intelligence. For meeting check-in and other scenes that need to take into account both recognition accuracy and response time. The conference sign in system developed in this paper is based on clmtrackr library to realize pure front-end live detection, and based on RetinaFace algorithm to realize face position detection. The optimized FaceNet algorithm realizes the extraction of 128 dimensional feature vectors of the face. In view of the problem that the original Inception of FaceNet is easy to cause the feature map to be too thick, the improved MobilenetV1 model is used to improve the speed of face recognition. The data set test results in LFW show that the average recognition speed is increased to 0.003 seconds per frame, which is 0.001 milliseconds less than that of the ordinary model per frame, and the performance is improved by 25.0%. At the same time, the accuracy can reach 97.3%.