Research On Improved CamShift Algorithm Based On Adaptive Color Feature

The detection and tracking of moving object is a valuable task for both theoretical research and practical usage, which combines theoretical knowledge from many different fields, such as image processing, computer vision and pattern recognition. This task can be widely used in national defense, the national economy construction and so on. However, the present application shows that the existing object detection and tracking technology is far from reaching practical application requirements. The key question is how to realize the tracking initialization and how to adaptively select the target features to track continuously. For this purpose, this paper focuses on reaching and discussing how to combine the detection and tracking algorithm. The major contents are listed as follows.First of all, according to the needs of the initialization for tracking algorithm, a detection algorithm based on Gaussian mixture distribution model are put forward in this thesis. By combing the detection algorithm and CamShift algorithm based on color feature, CamShift algorithm can automatically track moving object.Secondly, if the whole target is used as the initial search window of CamShift algorithm directly, the color feature of background will influence the tracking easily because of the complicate color feature of object. So an improved method is presented in order to solve the problem as below. Firstly, the object is divided into sub-regions. Then, the algorithm will compare the color similarity between the object sub-regions and background neighborhood. Finally, a sub-region will be determined as the initial search window. In this process, Bhattacharyya is used as the compared distance. Meanwhile, the color similarity will be compared real time in the tracking process to improve the real-time and robustness of the algorithm.Finally, we analyze the experimental result of improved method under the Visual C++ development environment and OpenCV libs. The experimental result shows that the tracking results have been improved obviously, proves the effectiveness of the improved algorithm.