Research On The Vision Tracking Algorithm Of Motive Object In Complex Scene

With the rapid development of science and technology, the cost of video acquisition is becoming more and more cheap, and the application of image processing is wider and wider. Benefiting from the a large number of image analysis algorithms, understanding of the environment via computer vision has gradually become reality. Like most important branches of computer vision, the main purpose of target tracking is to find the position of moving objects in image sequences, and obtain the motion parameters and position trajectories, and provide an important approach to the high-level application. There are a mount of work being taken in several domestic and foreign institutes, and a number of achievements have been got in recent years. However, there are still various issues in this field because of the complex and uncertain environment.As to solve the problems in target tracking, this paper firstly investigates the target and background characteristics of the complex scene. And the object description method and tracking feature selection criteria are discussed in the occlusion scene and fast motion scene which is on the basis of the general representation method of target. Finally, two tracking algorithms in complex scenes are presented by extracting appropriate features.For the problem of tracking in the occlusion scene, the scale invariant feature SURF operator is employed to detect the object by comparing with other local feature operator in which the Euclidean distance based clustering method is used to find the target potential position. Nevertheless, the original SURF algorithm searches feature points from the whole image that it is time consuming and to generate some mistakes. With regard to improving the speed and accuracy of target tracking, this paper proposes an improved algorithm based on Kalman filtering which is used to predict the target position. Thus, it narrows the scope of feature extraction and matching, as well as reduces the interference of complex background. In addition, it is of less time consumption. As shown in simulation results, the proposed method not only achieves good real-time performance and the matching accuracy but also enhanced the robustness of system by solving the problem of shape variation.For the problem of real-time tracking, this paper chooses Haar-like feature to describe the target, which represents the gray information of image. Although the Haar-like feature is easy to extract, it is of huge amount. Thus, certain training method or dimension reduction method is required to improve the tracking speed. This paper uses the compression tracking algorithm to extract multi-scale and high-dimensional features. This algorithm not only retains most information of original features, but also improves the speed of the tacking system. As a newly proposed approach, it is still of some problems like the drift. In order to improve the robustness of this algorithm, this paper proposes an improved algorithm based on the template matching method, which uses the gray histogram fitting results to determine the threshold. The experimental results show that the improved algorithm can solve the problem effectively and achieve no decrease of the tracking speed.Finally, all work in the research some deficiencies of the proposed algorithms are summarized respectively. The future considerations are also presented in detail.