Research On Dynamic Face Recognition And Retrieval In Surveillance Video

This paper mainly focuses on the research of dynamic face recognition algorithm for surveillance video data.With the development of computer technology and video imaging technology,surveillance cameras can be seen everywhere in various public places.The surveillance cameras will save the monitored video images to the hard disk for future search.The search content is mainly to find people.The current main search method is still manual.This is obviously a bit of a problem for the video images piled up in the hard disk.Therefore,it is necessary to use the computer to automatically search to improve efficiency.The video face recognition algorithm plays an important role in the search.The video data has a serious impact on the recognition effect due to its low resolution and various light occlusions,and the existing video face recognition algorithm is not enough to extract the spatial features of video data and to make weak use of video data timing information.Aiming at the existing algorithms that do not make full use of the spatial features and temporal information of video data,this paper proposes a video face recognition algorithm based on video data feature embedding.For the spatial features of video face data,convolution neural network with strong nonlinear feature expression ability is used to extract the spatial features.For the temporal information of the video face data,the algorithm extracts the temporal information of the video data while extracting the spatial features of the video data through the temporal shift module.The extracted temporal information is fully fused with the extracted spatial features by convolution neural network.Finally,the static face features extracted from the fused video face data are integrated by the temporal integration module proposed in this paper,and an embedded feature of the video data is obtained.For the extracted features of video face data,this paper updates the model parameters by metric learning.In this paper,the proposed video face recognition algorithm embedded with video data features is verified on the COX video face data set,and the method in this paper is verified in the video-to-video scene,in V2-V1,V3-V2,V1-V2 and V2-V3 have achieved good results in several experiment cases.Further,the mask data of COX dataset were expanded to simulate the situation of COVID-19 wearing masks,and the experiments were carried out on the expanded data set to further demonstrate the effectiveness of the algorithm.