基于Mean Shift的抗遮挡运动目标跟踪算法  

Anti-occlusion Motive Target Tracking Arithmetic Based on Mean Shift

作　　者：梁静[1] 支琤[1] 周军[1] 

机构地区：[1]上海交通大学电子工程系图像通信与信息处理研究所,上海200240

出　　处：《电视技术》2008年第12期82-85,共4页Video Engineering

基　　金：国家杰出青年自然科学基金(60625103;60702044)

摘　　要：采用Mean Shift和Kalman滤波器相结合来处理动态背景下目标跟踪问题。首先利用Kalman滤波器进行预估计获得每帧Mean Shift算法的起始位置。由图像差分法得到物体轮廓,同时定义了相似因子判断物体是否发生遮挡。当发生遮挡时,根据物体运动状态不同,对颜色信息和运动信息分别赋予不同权值来预测物体在当前帧的位置并作为下一帧预测的起点。此时,目标位置的线性预测替代了Kalman滤波器的作用。实验证明,新算法可实现对快速运动目标的跟踪,对遮挡也有很好的稳健性。An algorithm for tracking fast motion objects under dynamic backgrounds is proposed, which combines Mean Shift and Kalman filter. At first, the Kalman filter is used to get the predicted starting position of Mean Shift in every frame. And then, the contour is got with the image difference method. At the same time, the similarity factor is defined to determine whether the tracking target is occluded. When the occlusion happened, the color information and motion information are assigned with different weight values to predict the position of the tracking target in the current frame and set it as the start position of next frame ac- cording to different motion state of the target. In such situation, the Kalman prediction is substituted by the linear prediction of the target. The test results show that this algorithm can track the fast-moving target well and also has better robust for occlusion.

关 键 词：目标跟踪 Mean SHIFT算法 KALMAN滤波器 相似因子 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]