A Study Of Object Tracking Algorithm Based On Level Set Method

Being one of the most important topic in computer vision and pattern recognition, object tracking has been dramatically developed in recent years. It possesses broad requirement and is applied in the area of video monitoring, human-computer interaction, medical diagnosis, weather analysis, security monitoring and missile guiding, etc. Finding the interested object region in video frames is the main purpose of object tracking, and similarly partition the interested areas from video frames into is what the segmentation does. Then, considering the close relationship of object tracking and segmentation, incorporation the segmentation into object tracking has gained a lot of attention of scholars. Level set methods have been well developed and widely applied in the field of segmentation. How to apply the level set methods into object tracking is the main content of this article.On the basis of current research situation, we categorize the object tracking algorithm based on level set into two ideas: one is tracking by stepwise tracking and segmentation, and another is tracking by mixing tracking with segmentation. As the former one, the segmentation module and the tracking module are independent of each other. The tracking module determines the candidate target areas, and the segmentation module segments the candidate target areas. The result of segmentation can be used to improve the tracking result, then the object tracking is accomplished by combining tracking with segmentation module. The tracking module is integrated into segmentation module in the latter ideas. The bound variable of object description is added to the evolution function of level set, therefore the result of tracking is obtaining by segmentation directly.This paper firstly introduces the theoretical fundamental of level set based segmentation. The theoretical principle of level set method has been introduced in detail, and the object tracking algorithm based on level set has also been summarized. Secondly, the theoretical principle of Particle Filter has been introduced in detail. According to the work of other researchers, a level set tracking algorithm based on shape constraint particle filter is proposed. According to the object shapes obtained by level set segmentations, color histogram is optimized, and shape similarity is constructed. Thenthe definition of particle filter weight is improved to make tracking result more accurate. The results of experiment simulation, compared with particle filter, the level set tracking algorithm based on particle filter purposed by other researchers, the level set tracking algorithm based on shape constraint particle filter, show that the proposed one has a better result than others. Finally, the theoretical principle of Gaussian Mixture Model has been introduced in detail, and a level set method object tracking algorithm based on Gaussian Mixture Model proposed by other researchers is introduced and analyzed. According to the analysis of this algorithm, an improvement method of this algorithm is proposed. By modifying the inner bound item of the level set evolution function, the result and Efficiency of segmentation and tracking is optimized. Through the results of experiment simulation, compared with the level set method object tracking algorithm based on Gaussian Mixture Model, proposed by other researchers, shows that the proposed one has a better result than another.