Research On Intelligent Systems Based On Face Recognition Technology And Edge Computing Technology

With the rapid development of deep learning field,the research and application of face recognition and expression recognition technology has received increasing attention in the age of big data.In the area of security monitoring and man-machine interaction,more and more related applications appeared.The common face recognition systems work in offline way or networking way so far.The algorithm in offline face recognition system has fast computing speed,but has low recognition accuracy.The algorithm in networking face recognition system has high recognition accuracy,but takes too much time in computing and transmission.To solve above problems,we focuses on designing two face recognition intelligent systems in this paper.Firstly,using deep face recognition neural network and network model compression technology,we designed an offline face recognitionsystem with fast computing speed and high recognition accuracy on a small embedded board.The system takes only about 200ms to extract the deep feature of a face picture without GPU acceleration,and makes great contribution to promote the offline face recognition system based on deep learning.Secondly,using face recognition algorithm,edge computing technology and Q-Learning reinforcement learning algorithm,we designed a networking intelligent video surveillance system.The system process the video surveillance data intelligently with the help of recognition algorithms in front modules and back modules and decision-making algorithm in decision-maker module.So the intelligent system can solve the problem of occupying too many storage and computing resources and response slowly in traditional surveillance system.In the end,we introduce some work about facial expression recognition in this article.We propose a face expression recognition algorithm based on multiple local feature fusion.The algorithm improves the recognition accuracy and solves the problems of poor recognition performance using traditional single expression feature.