Person Re-Identification Based On Small Groups Context

With the rapid development of economic construction and Urbanization process,Video surveillance network coverage is increasing day by day.And the Intelligent video surveillance demand of urban security is increasing rapidly.on this background,a demand of re-identifying pedestrian in surveillance video by using computer vision technology emerged,which is termed as person re-identification problem.More precisely,person re-identification is to identify whether a detected person has emerged before in the surveillance camera network.With the increase of technology development and application requirements,person re-identification is increasingly becoming a hot research spot in the computer vision community.Taking into account the influence of different cameras under the monitoring network on the appearance of pedestrian images,and in public places,it is generally carried out in small groups.This paper considers the use of small groups as a robust cue for person re-identification.Combined with the latest computer vision research theory about the person re-identification problem,we implement in-depth research for some specific key problems.The exciting parts of our article are four aspects:(1)In the aspect of small groups contextual information mining,Small group velocity information is obtained by particle motion trajectories.According to particle motion trajectories and the social cognitive rules,similarity matrix will be constructed between particle trajectories.Then we use the similarity matrix to distinguish different moving groups by means of spectral clustering method in the scene.(2)In the feature representation,we consider the feature of small groups as a contextual cue for the person feature.For the situation of small groups in the presence of different cameras,such as space position changes,occlusion,small group feature based on rectangular rings is designed to represent small group image.And we design small group similaritycalculation method to improve the accuracy of small group matching.(3)In distance measure,traditional distance functions are not proper as they do not consider the distribution of samples.This paper formulate person re-identification as a relative distance comparison learning problem in order to learn the optimal similarity measure between a pair of person images.This model is formulated to maximize the likelihood of a pair of true matches having a relatively smaller distance than that of a wrong match pair in a soft discriminant manner.And an iterative optimization algorithm is designed.Moreover,this paper further develop an ensemble model.(4)In terms of re-identity strategies,considering the small group contextual information and the re-identity model,this paper propose three strategies based on small group context.This paper make full use of small groups of contextual information and apply it to the person re-identification area.The experiment shows that our method can effectively improve the accuracy of person re-identification.