Person Reidentification By HMM

In recent years, with the constant reduction of camera production cost and the importance people payattach to public security, video surveillance system has achieved rapid development and is widely appliedin various places of the city, such as airports, banks, supermarkets, and other government organizations.Therefore, identifying a certain person in multi-camera surveillance system and determining whether he orshe appeared in a given time or a given place, which is often called person reidentification or peoplereidentification, has become a very important issue. However, relying solely on security personnel tomonitor the changes in the camera24hours a day takes a lot of manpower and material resources, whichhas made person reidentification task become hot issue in computer vision field.Person reidentification is the key to continuous objects tracking in multi-camera environment andvideo object retrieval in large-scale video data. This paper presents a novel method for personreidentification by using feature extracted from multiple frames of the person object to train hidden Markovmodel, and then identification is based on those well-trained models. Normally, the difficulty of extractingfeatures lies in the computational complexity and high dimensionality, but the proposed feature extractionmethod has already overcome these shortcomings. Feature extraction algorithm’s time complexity is linearcost, and it also can accurately represent the apparent characteristics of object. In the modeling stage,complex feature requires the algorithm to select a specific feature channel, while hidden Markov modelfuses multiple features and it also effectively combines the structural constraints of human objects.Finally, experiments on two publicly challenging datasets demonstrate the effectiveness and robustness ofthe proposed method, the experimental results show that the proposed methods are robustness againstillumination changes, viewpoint change and low-resolution.