Face Detection And Recognition Algorithm Based On Deep Learning For Video Surveillance

As an intrinsic biological property,human face has strong stability and individual difference.The properties of universality?uniqueness and easy to collect of human face have unique advantages in video surveillance system authentication.This thesis mainly discuss face detection and recognition algorithm based on deep learning for video surveillance.In general,face images captured from video surveillance differed in gesture and always with low resolution which poses challenge to face detection and face recognition.Recently,with the great success in the fields of language recognition and image classification,the strong characterization of deep nonlinear network structure has been paid more and more attention.It has been applied to the field of detection and recognition gradually and reached the application level of real-time and accuracy.This thesis analyzes the difficulties of face detection and recognition in video surveillance.Referring to the current application of deep learning in face detection and recognition and related fields,this thesis proposes a face detection in video surveillance and rapid face recognition in massive databases.The main research work of this thesis is as follows:(1)This thesis summarizes the advantages and disadvantages of traditional face detection and recognition methods,analyzes the advantages of deep learning theory in face detection and recognition,and the difficulties in the application of surveillance video.(2)Aiming at the large amount of monitoring video data,a cascaded network is improved for face detection,which improves the speed of face detection.(3)Aiming at the fact that the current face recognition algorithm has low retrieval efficiency in practical applications,a residual network is constructed,and some appropriate face loss function are proposed.An algorithm for extracting facial binary features is proposed,combined with K-Nearest Neighbors Searches algorithm,which improves the efficiency of face retrieval.(4)Set up a software operating platform to organize detection and recognition algorithms and migrate them to the platform,and perform overall testing of the algorithms on the platform.