Study On The Moving Target Tracking Based On Mean Shift

As in the subaltern studies research area of the computer vision, moving target tracking is a technology which can get the target location from the video on time, and track moving target automatically. It provides valuable information for advanced processing such as event monitoring, behavior understanding and description. It has a very wide range of applications in the smart monitoring, human-computer interaction, image compression and three-dimensional reconstruction. Target tracking requires accuracy, robustness and real-time. The Mean Shift algorithm uses the nonparametric density estimation technology to climb to the density maxima along the direction of density gradient quickly and efficiently. Mean-shift based on visual target tracking algorithm is simple, efficient and easy for module implement, and it can deal with the complex cases such as the edge noise, interfered objects and partial occlusions.This paper introduces the Mean Shift algorithm as well as its detail deduction, and it also proves the convergence of algorithm reliably. At the same time, this paper introduces the traditional target tracking algorithm based on Mean Shift. It points out the advantages and disadvantages of the traditional target tracking algorithm, and makes the following improvements for its shortcomings:First, this paper combines the Mean Shift with Kalman filter, which is used to forecast possible position of target, then Mean Shift searches the real position near the possible area. The algorithm has good effect on fast moving target tracking and occlusion target tracking.Second, this paper combines the Mean Shift and target detection algorithm. Target detection algorithm , extract the region of moving target in the complex background, can provide the target's size for the Mean Shift algorithm to achieve the adaptive update of Mean Shift's template and kernel bandwidth.Third, this paper research for the target detection and shadow detection algorithm. On the one hand, a new concise studying mode in background model is proposed to reduces memory amount of the nonparametric model; on the other hand, this paper proposes an adaptive shadow suppression algorithm, which makes the shadow detection applied to a variety of different scenes.The key algorithms of the paper are realized using OpenCV in the VC++6.0 programmed environment. The experimental results indicate that the algorithms work well, fast and stably.