Optimization of Face Recognition Systems for Implementation in Embedded Systems

The article considers the problem of identifying a person by face. Particular attention is paid to primary face detection algorithms. A significant acceleration of the algorithm is shown when using the YuNet network for a large number of individuals. In addition, motion detection without neural network algorithms is proposed, which also speeds up the processing. The results of the work of various classifiers are presented, and their comparative analysis is performed. It was found that the use of the ArcFace loss function provides better accuracy of classifiers. The possibilities of processing video sequences are demonstrated. The developed algorithms have high accuracy and they are characterized by high performance.