Research On Moving Object Tracking Algorithm For Intelligent Visual Surveillance

Video-based motion object tracking has long been a very important and active research topic in the field of computer vision, image processing and pattern recognition. In recent years, one of its main applications is video surveillance sysytem. In the video surveillance system, it is a challenge to track the target accurately in complex environment and different conditions, a diffcult problem to solve in practical application currently. Regarding to the different scenarios of moving object tracking, Mean Shift-based motion tracking algorithm is chosed as core of our study. Main work of the paper as follows:Video-based tracking research status is described and various off video-based motion tracking algorithms are classified. Meanwhile, advantage and disadvantage of various motion tracking algorithms are pointed out, and feature choice and target representation of them are also discussed.Paper focuses on Mean Shift algorithm and its application on motion tracking. In the video tracking process, the target area and the corresponding histogram are usually established in the first frame by user. In the subsequent frames, the best target candidate areas are searched iteratively by Mean Shift algorithm based on the Bhattacharyya function. Mean-Shift method can be peformed in real time and is robust for partial occlusion and distortion.However,when the target is similar to the background,it is difficult to distinguish the bakground and leads to fail in tracking.Based on the above, the basic idea of Mean Shift tracking algorithm and its advant ages and disadvantages are described. In Mean Shift tracking algorithm, the normalized weighted color historgtam model is generally used as a way to describe the target, it is a descrete probability density estimation of. target color distrbution. The statistical characteristics of the overall target is described using color histogram, but it does not include the spatial information of target. So, the way of blocking the target region is used and then the Bhattacharyya function of every block is calculated.The block whose Bhattacharyya function is the max is chosed out and its location is made as the location of target in next frame. This method increases the amount of space information distribution to rich the characteristics of object model describing. Also it has a high capacity of recognition and robustness to partial occlusion of the target. Experimental results show that the algorithm has a good accuracy and robustness to track the motion object effectively.It is used only one single color feature to representate the tracking target in Mean Shift tracking algorithm, so target tracking in complex secens often leads to failure. In this paper, the joint histogram of color and texture and the Kalman filter prediction is introduced to predict the approximate location of moving target's center point in current frame, then the point location is made as starting point of Mean Shift iteration. And then Mean Shift Algorithm is used to find the true position of target in the neighborhood of the point to solve the occlusion problem, Experiment results show that the algorith can achieve a good tracking result when the pose, illumination of moving object get changed and especially robust to occlusion.