Research On Person Identification In Survalliance System

With the popularization of surveillance system,the task of analysing the data generated by the surveillance system becoming more and more burdensome.As person becomes one of the most important monitoring objects,researchers dedicate themselves to the fields of person detection and recognition.In intelligent surveillance system,person identification is the fundamental component.In this paper,we investigate the person identification problem,and propose a framework for person re-identification and a framework for unconstrained face recognition.Person re-identification aim at associating the same person across different surveillance scene at different time.It identify a person by his/her appearance.In this work,a reranking framework based on manifold is proposed.First,the reranking framework utilize the low-level feature that overlooked by the conventional person re-identification algorithm to generate the manifolds on different modalities.Then,the manifolds on different modalities are fused into one manifold.Finally,the distances of probe image and images in the gallery set are computed on the fused manifold.In the experiments,4 person re-identification algorithms are integrated into the framework,and we test the framework on 2 public datasets.The results demonstrate that the proposed framework can improve the conventional person re-identification algorithms remarkablely.Unconstrained face recognition aims at identify a person when his/her head pose is unconstrained.It identify a person by his/her face.In this paper,we improve a unconstrained face detection algorithm,and combined it with a tracking algorithm.Then we get a tracking-based face detection algorithm.Based on this tracking-based algorithm,we propose a unconstrained face recognition system.