Research On Human Detection And Tracking In Infrared Image Sequence

The human detection and tracking is a hot research topic in the field of computer vision. Infrared imaging system has a stronger visual ability in the darkness and smoke environment comparing with the traditional optical imaging system, almost can realize working without a pause. Therefore, the research of the human detection and tracking in infrared image sequence has the value of practical application.In this thesis, the candidate target extraction algorithm, the human recognition and classification algorithm of the human detection and the human tracking algorithm in infrared image sequence are studied on base of the introduction to the analysis of the characteristics of the infrared image.A new background model construction algorithm based on the hypothesis of background pixel with maximum probability is proposed. Firstly, a fixed pixel in several infrared images are classified based on the distribution of its gray level. Then, the background class is defined as the class with the maximum pixels. Finally, the Gaussian model is used to describe the gray level of the pixels and their neighborhood. The experimental results show that the new background modeling method can accurately construct the background of the image sequence and has robustness to various situation, the background subtraction can accurately extract candidate target.Aiming at the recognition and classification of the candidate human target, a new edge feature based on grid partition and local description is proposed. The shape features, the duty ratio features of the target is extracted and these three features are extracted to describe the human target according to the thinking of global to local. The cascade classifier with three levels is used to recognize and classify the candidate targets, the experimental results show that the new human detection algorithm is able to gain good results for different databases and all kinds of situation.The particle filter algorithm is used to realize the tracking of human in infrared image sequences and a new human feature based on the recombination of the local edge features is proposed to overcome the inaccurate tracking result caused by shelter. Firstly, the local edge features are extracted. Then, all the adjacent sub-region’s edge features are extracted to form the new human. Finally, the particle filter is combined with the new human feature to realize the human tracking. The experimental results show that this human tracking algorithm has the high accuracy not only under the normal state but also under the situation of shelter.