Tracking And Matching In Multiple Non-overlapping Cameras

We will explain the important parts which are background modeling, tracking, matching in the thesis.A sequence comprises time and space information. If we want to get valid back ground model, we should combine the temporal and spatial information to improve the computing speed and resource consumption. Our approach takes full advantage of the temporal and spatial information, and removes the redundant data. We take two steps to updat: the first step updat the parameters roughly range, the second step refine them. It has more advantages than the classical back ground model, such as GOM, KDP. In this way, the method is fast, simple, effective, no initialized limitation, and robust. It has practical applications, either in hardware design or software design.Targets'tracking is an important content in our system. Through studying particle filter, we improve the observation likelihood, and get a robust tacking result. At the same time, using the tracking feature LBP,HSV,EOH in the change illumination condition, we get the different tracking results. So we can fuse some tracking features to handle diverse conditions.We proposes an fast background modeling method with the improved particle filter that fuses different tracking features. We get different matching results by adopting statistical histogram, temporal and spatial relation, and BTF as three matching foundations. Finally we successfully realize tacking and matching in multiple non-overlapping cameras system.