Persion Re-identification Research For Video Surveillance Application

In recent years,cite safety has been paid more and more attention because of its' close relationship with our daily life.The video surveillance system has been constantly improved to ensure the safety of the city.The coverage rate of the video surveillance camera has been continuously improved,leading to that more attention has been payed to the intelligent behavior analysis based on video surveillance network.With the extensive layout and development of video surveillance,person re-identification as an important research direction in the analysis of intelligent behavior has been payed more and more attention.Person re-identification is a problem that identify the same person from different surveillance cameras in video surveillance network.Although a large number of studies have focused on the problem of person re-identification,it still remans very challenging.View variation,illumination change and occlusion are all the challenging problems in person re-identification.With the goal of improving the accuracy and efficiency of person re-identification,we propose a person re-identification framework using cascaded filter based on summarizing and analyzing the existing methods of person re-identification.We also propose a person re-identification algorithm based on the fusion of biometrics and appearance features.These two methods have good performance in improving the accuracy and efficiency of the person re-identification algorithm.At the same time,in view of the application requirement of Shanghai city video surveillance,based on the above two algorithms,we design a software system for person re-identification,which is used in the video surveillance system.It plays an important role in intelligent behavior analysis.Feature fusion is proved to be effective in the problem of person re-identification.The distinctiveness and reliability of feature can be improved by fusing multiple features.Since single feature is not powerful enough to tell every subtle differences between pictures because of view variation,illumination change and occlusion,commonly the fusion feature can performs better than single features.In this paper,We propose a cascade filter person re-identification framework,making full use of information provided by features.We address how to take advantages of different ranking results yield by using different features to alleviate the influence of the error determination when the ranking results of different features are not consistent.In addition,by using this filtering mechanism,we can decrease the number of the image samples,which can reduce the computation of feature extraction and matching,improving the efficiency of the method by reducing the time cost.The feature used in person re-identification can be divided into two categories.One is the appearance feature.At present,most of the person re-identification methods are based on the appearance feature.Another is biometric.For example,we can use the face feature in the pedestrian images or walking gesture to distinguish different individuals.In daily life,we always distinguish different individuals by the face,which is because that the same person's face in most cases will be relatively stable and the face feature is an important discriminative characteristic for person re-identification.But in the early stage of development of video surveillance system,because of the low resolution of video surveillance cameras,it is very difficult to extract facial information from the pedestrain image.However,HD cameras are increasingly used in surveillance systems,it is possible to get effective face information.Although the camera view change and high position of the camera will give a negative impact on the application of facial features in the person re-identification,we can make the face feature of pedestrian image play a role in the person re-identification method by using our framework which fusion biometrics and appearance features for person re-identification.This framework make use of the information of the ideal human face image and give up the unsatisfactory face image,making the most discriminative face feature of human play a role in the person re-identification system.We also choose complementary appearance features for person re-identification,fusing the appearance features with face feature,making full use of information of different features and improving the performance of person re-identification.Finally,according to the demand of video surveillance system for intelligent behavior analysis,we integrate the algorithm into the software,designing and constructing a person re-identification software.The algorithm is integrated into the monitoring scenarios,making person re-identification play a role in video surveillance system.