Joint Temporal-Spatial Information And Common Network Consistency Constraint For Person Re-identification

Since entering the 21 st century,public safety incidents occur frequently,and have posed a tremendous threat to the life and property safety of people in various countries.This phenomenon have also drawn great attention from the people and governments in various countries.Urban safety is paid more and more attention as an issue because of its' close relationship with our daily life.With the rapid development of image acquisition technology and large-scale data storage technology,video surveillance has become an important way for governments of all countries to raise public safety management standards.Large-scale camera networks are increasingly deployed in public places.In this context,the issue of person re-recognition has drawn more and more attention in automatic video surveillance as an important research direction in intelligent video analysis.Person re-recognition is a task that identifing the same person from different surveillance cameras in a multi-camera video surveillance network.Although a great deal of research has focused on person re-recognition,the problem still faces a very serious challenge.Based on the analysis of the existing methods of person re-recognition,aiming at improving the accuracy of person re-recognition algorithm and improving the efficiency of the camera person re-recognition algorithm,this paper proposes a simple multi-camera network person re-recognition algorithm framework.In the framework,a method of defining spatial-temporal information is proposed,which uses a combination of visual information and spatio-temporal information algorithm and a network consistency strategy.We take full advantage of the information provided by the various characteristics of network.When the result of person rerecognition given by visual information is misjudged or contradictory,the network information can assist in troubleshooting these mistakes.With single use of the two methods in the framework,both have a good performance,and then with organic combination of the two method,performance improvement is even more apparent.At the same time,the part of the network coherency strategy in the framework will be statutory to integer programming.In this framework,integer programming is the glue that integrates visual information,space-time information,and network consistency strategy into a single framework.In order to give full play to the advantages of the framework,this paper discusses in detail the solutions to the integer programming problem of the network consistency strategy and gives the solution.In addition,taking into account the diversity of actual monitoring scenarios,aiming at the complicated scene with complicated camera network topology and irregular pedestrian distribution,this paper proposes a method to directly join the camera topology information when calculating the similarity score between persons,improving the reliability.