Research On Target Tracking Algorithm Based On Color Name

Target tracking has been a major research in the field of computer vision research. Target tracking is very significant in real life, for example, it is widely used in intelligent transport systems, video surveillance systems, human-computer interaction, video search, video compression, 3D reconstruction, virtual reality. Although the target tracking technology has gained great progress in the past few decades, it still has some unresolved issues, such as lighting changes, occlusion, foggy day, rain, snow days, changes in the size of the target, background clutter and real-time requirements.etc.The adaptive color attributes for real-time visual tracking(CN) algorithm is a kind of algorithm basing on discriminant target tracking algorithm, and it uses the efficient regular least squares method classifier to replace the traditional SVM(Support Vector Machine) classifier to classify, adhibits the cyclic matrix and Fourier theory to dense samples, adopts the color features from CN space and gray values on behalf of the feature of the target. It exploits the method of principal component analysis to reduce high-dimensional color data, which can simplify the computational complexity of the tracking target. The final, the parameter of the classifier will be updated in real time, which considers all previous frames. The performance of the CN tracking algorithm is better than most current tracking algorithm, especially in the real-time apspect. Thereby obtaining scholars widespread concern and attention. However, for one thing, the CN algorithm tracks the target only depending on its color characteristics; for another thing, the algorithm only determines the next frame's search scope and the center of Gaussian filter which uses to image preprocessing depending on the center of the current frame target. Since these defects would cause the algorithm deflect from or lose the target on some sequence images.In order to understand the current tracking algorithm and further improve the performance of the CN algorithm, the main work of this thesis includes:(1) Classify and analyze the existing target tracking algorithm, and analyze the principle of the CN algorithm with more detail.(2) Because the CN algorithm only uses the color feature to track, it is easy for this algorithm to lose target in the case that the target is blocked or the color of background and target are similar in a series of images. To overcome the problem, we propose that it should combine the CN algorithm with the algorithm which uses to extract the edge feature from the color feature to track the target. The Monte-Carlo simulation results show that the CN algorithm based on the Laplace of Gaussian can effectively improve the tracking effect.(3) On the one hand, because the CN algorithm only uses the center position of the current frame target to determine the search scope of the target of the next frame, it is easy for this algorithm to lose target in the case that the target is not in the the search scope of the target. On the other hand, the CN algorithm only uses the center position of the current frame target to determine the center of Gaussian filter which uses to image preprocessing of the next frame. Facing this problem, we propose that it will be more appropriate to shift the center position of the current frame target inertially to determine the search scope and the center of Gaussian filter of the next frame target according to the shifted center. After blending the inertial drift thought into CN algorithm, which will be more accurate to combine the center position and the size of the target to locate the search scope of the target. The experimental results show that the algorithm is not only better than CN algorithm but also better than some good target tracking algorithms produced in recent years. This algorithm has higher tracking precision and its average tracking speed is up to the standard requirement of the video.