Research On Face Recognition Method Based On Video Surveillance System

Today,with the development of science & technology,face detection technology in the field of computer vision & pattern recognition has been applied to adapt to a variety of model scenarios.In the public security departments in the fight against terrorism,the maintenance of order in the practical work,in particular,it has an excellent role.For example: For large scenes of video surveillance in any one person identification,identify its identity immediately when a person into a special area.When in the study of face recognition technology,researchers carried out a detailed classification of the subject,they can be divided into human face detection,location tracking,sample identification & so on.I have been working in related areas,I need to monitor a large number of surveillance cameras taken by public security matters in Pudong New Area,& to monitor the object of the face detection,tracking,compare with the database of criminal data if it is necessary.In order to be able to do my own work,& to increase knowledge reserves,I chose this subject for research,for the selected project content,in this paper,we consider the following aspects:(1)First of all,this paper expounds the significance of face recognition & its wide application prospect,& makes a brief analysis of the present situation of this research.The advantages & disadvantages of several common face color models related to skin color are briefly introduced,& the face detection methods based on skin color & image likelihood are introduced & analyzed in detail.Because this is the method used in this study,& then simply describes the basic flow & steps of common face detection technology,& analyzes & describes the face recognition & common interference factors in complex scenes.(2)In this paper,we focus on the hotspots of current face tracking technology,& introduce the commonly used algorithms & techniques.The Mean Shift algorithm is introduced in this paper.The basic principle & the main implementation steps are introduced in detail.Meanwhile,the gain of the algorithm & the shortcomings of some specific scenarios are also presented.(3)Based on the research results & experimental data,this paper proposes a block mean shift algorithm based on target real-time adaptive updating,which combines LBP(local binary pattern)with histogram to obtain local texture features.The real-time adaptive updating of the target improves the accuracy of the target tracking algorithm,Furthermore,the local feature of the target can be further enhanced by segmenting the target,which improves the robustness of the algorithm under the skin-color background.Finally,the paper analyzes the shortcomings of using Mean Shift algorithm.Such as it can not recover the tracking in a short time for the reason of the target is lost suddenly,& it can be combined with frame difference method or eye module assistant method to improve its effect.(4)Based on successful face detection,in this paper,we introduce & analyze several commonly used face recognition algorithms based on feature subspace,& make full use of the advantages of the subspace method.At the same time,based on the recognition of the defects in the samples extracted by the single training sample feature,an improved 2D-LDA algorithm based on two-dimensional image matrix is proposed.In this method,we first need to increase the sample variables & extend the sample size to 12 by using the method of general sliding window & rotation transformation.Then,we extract the block features from the obtained sample size & strengthen the local variables feature.After a series of conversion work,the algorithm proposed in this paper,although a little less,improve the accuracy of face recognition effectively.(5)A video surveillance face tracking system is designed & experimented with its own work.