The Research On Single-Camera Tracking Based On Texture Character

Object tracking, which is widely applied in intelligent surveillance fields, is abstracting objects in videos in real time and forecasting their locations and velocities. The research on single-camera tracking is the basis of the research on object tracking. This paper researches single-camera single-object and multiple-object tracking. The innovated point of the research is that the histogram texture character of objects is analyzed and is used to abstract and to identify objects in the process of single-object or multiple-object tracking.At first, this paper validates the method of using color histogram characters to abstract foregrounds, proposes a fast texture abstraction method, making use of the color histogram of foreground templates and backgrounds, and the Bayesian theory to compute the foreground and the background probability of pixels, using the region grow and the foreground assimilation algorithm to abstract foreground regions. The experimental result shows that the proposed method uses less time to the similar method in the literature and its abstracting effect do not fall out.Following, this paper improves on the window tracking algorithm, analyzes the histogram character of the moving object, and based on it processes the proposed texture abstract method at the current quality point to abstract objects. The experimental result shows that the proposed algorithm not only catch the object when it moves, but also abstract the object based on its texture character when it hold still. The proposed algorithm availably solves the problem of non-integrality and reliant to movements in abstracting objects. But the lack of it is that it is greatly influenced by the texture abstraction algorithm and is difficult to distinguish the foreground and the background when the background becomes complex.Finally, this paper uses the color histogram character of objects to match and to track them in the process of single-camera multiple-object tracking. The experimental result shows that the proposed algorithm can reach the anticipant effect of multiple-object tracking and solve the object-clustering problem in a certain extent.