Research And Implementation Of Person Re-identification System Based On Multi-camera Network

With the deployment of survelliance cameras in quantity,the scene of the intelligent video analysis is in transition form single camera to multi-camera.The multi-camera network brings more monitoring information to people,and also brings differences between frames of different cameras,as well as mass video data to be processed.The popular research directions cover the problems of getting effective information from the monitoring environment of the multi-camera network,and meeting the various monitoring requirement.This thesis focuses on person re-identification in multi-camera network.Firstly,research on the feature engineering of person re-identification is done by us.Methods of pedestrain feature extration by handwork,and methods by deep learning are investigated and analyzed.The selection of deep learning models is studied and the corresponding experiments are done by us.Secondly,based on the previous deep leaning method of feature extration,the ways of building index on them in the scene of mass data are summarized,and LIRE,a open source CBIR framework,is extended by us.A solution of offline pedestrain retrieval system in distributed way is also provided.Thirdly,in order to use spatio-temporal information in the multi-camera network,the concept of 'the ring of encirclement to cameras' is proposed and the usage is provided.Finally,the key technologies of online person re-identification are studied and the complete solution of the online person re-identification system is provided.The major parts of the system are implemented and the experiments of them are done,in order to prove the good performance of it.