Face Recognition Algorithm In Surveillance Video

With the rapid development of technologies in computer vision,automatic face recognition technology,which improves the intelligentization degree and comprehensive processing capability of surveillance systems,is able to identify faces more quickly and accurately under uncontrolled conditions.However,because of the existence of difficulties,such as slow speed of video face feature extraction and differences between face images in surveillance video and database,it is challenging to apply face recognition technology widely in surveillance scenarios.This aims at exploring the ways to fuse the video face features efficiently and eliminate the impact from the differences between images on the recognition accuracy.The main achievements in the thesis are as follows.1.A feature fusion algorithm based on key face frame extraction in video is proposed.The key frames are extracted based on the cosine distance between the features and the identity weights,then those high quality face features in the key frames are selected for fusion.In this way,the speed of video feature extraction is improved and robustness of the fused features is enhanced at the same time.Experimental results show that the proposed method can effectively improve the speed and accuracy for video face recognition.2.A face recognition algorithm for surveillance video based on feature distance metric and dual branches structure is proposed.With a more refined probe loss function,the distance relationship between features is reconstructed by using the dual-branch structure to process different environments separately.Experimental results show that the proposed algorithm can eliminate the influence of differences between images on the recognition accuracy of surveillance videos.3.Another surveillance video face recognition algorithm based on uniform feature representation is also proposed.Based on the adversarial learning method,the unified feature expression is constructed by eliminating the source information of face features.Experimental results show that the proposed algorithm can improve the accuracy of surveillance video face recognition.4.A demonstration system for face recognition in surveillance scenarios,is built to perform a complete process of recognizing face identity from surveillance video.