Research On Target Tracking Algorithm In Natural Scence

In recent years, target tracking technology has been applied in more and more areas, such as video surveillance, auxiliary driving system, military navigation system, etc. With the widespread use of the target tracking technology, it has gradually become a research hotspot in the field of computer vision, and attracted the extensive research interest of scholars. At present, researchers have proposed a stream of target tracking algorithms, and made some achievement. But there are still some problems, such as the gesture change of non-rigid targets, environmental illumination change, occlusion, etc. Because the background in natural scene is more complex, they have a greater effect on the tracking performance. Therefore, this article focused on target tracking in the natural context, and the main research content is organized as follows:Firstly, the fundamental theory of target tracking is introduced. In this paper, the features commonly applied in target tracking technology have been introduced and analyzed, which consists of color feature, texture feature, gradient feature. Furthermore, the theories of the tracking algorithms used in the article are introduced, including the particle filter, Kalman filter and the mean shift algorithm.Secondly, in order to solve illumination and gesture change during the process of target tracking, a sparse representation tracking method was proposed based on local sensitive histogram. Local sensitive histogram features of multiple candidate targets were extracted, and sparse representation coefficient for each candidate target was calculated on the basis of template dictionary by using modified LI norm model. Moreover, the weight for each candidate target was calculated. The candidate target which had the largest weight was selected as tracking result. Experimental results demonstrated that the method can track the target accurately and effectively and has advantage in illumination and gesture change.Finally, the research proposed a traffic sign tracking method which combines Kalman filter with adaptive mean shift and definition evaluation method. First, the traffic sign detection method is applied to obtain target state in order to initialize the tracking algorithm. Then, the traffic sign is tracked by leveraging both Kalman filter and mean shift algorithm. In the tracking process, the Reblur algorithm is adopted to recognize the resolution of traffic sign. When the resolution reaches to the extent that can accurately identify the traffic sign, the traffic sign tracking method proposed stops traffic sign tracking and enters sign identification process. The experimental results show that the algorithm can well track traffic sign, and recognize the resolution of traffic sign.