Research on Face Matching Algorithms for Smart Classrooms

Face matching technology is a key part in the applications of face recognition technology. It can be used for classroom behavior analysis, and then improve the teaching effect. The paper introduces the algorithm structure and loss function of face detection algorithm based on MTCNN, and realizes face comparison based on FaceNet model. According to the different tasks of training data set and algorithm in different convolution layer, the diversity of face samples in training set is adjusted. Through MTCNN's sub network training, the weight coefficient of loss function and various thresholds are changed to improve the training effect and enhance the robustness of face detection. Considering the problems of multi person detection, face recognition illumination robustness, algorithm implementation cost and application real-time, this paper integrates MTCNN detection algorithm, FaceNet model and image similarity comparison algorithm, and implements the algorithm on LFW data set. Then, under the condition of specific face segmentation, the selection method of the best matching threshold is analyzed, and the results of Euclidean distance comparison of different faces under the best threshold are given. At last, it gets the recognition accuracy improved by 0.21% on the LFW dataset, which lays a technical foundation for the realization of classroom behavior analysis of students' location attendance.