基于改进的均值漂移算法的非刚性目标跟踪  

A Non-Rigid Object Tracking Algorithm Based on Improved Mean Shift

作　　者：赵瑶[1] 常发亮[1] 郝洪霆[1] 

机构地区：[1]山东大学控制科学与工程学院,山东济南250061

出　　处：《计算机工程与科学》2007年第12期71-73,81,共4页Computer Engineering & Science

基　　金：国家自然科学基金资助项目(60775023);山东省自然科学基金资助项目(Z2005G03)

摘　　要：针对现有的均值漂移算法不能适应非刚性目标的复杂运动情况,本文首先利用基于边缘的背景减方法去除背景干扰;然后利用GVF Snake技术提取出目标轮廓,结合目标轮廓改进了传统的颜色直方图;最后基于该颜色直方图结合卡尔曼滤波器或粒子滤波器改进了传统的均值漂移算法。实验表明,该算法可以实现快速的非刚性目标跟踪,对目标的不规则运动和严重遮挡有很好的鲁棒性。The mean shift algorithm is improved to deal with the poor performance of the current mean shift in tracking non-rigid objects with eomplex movements. First, distractions in the background are removed through background subtrac- tion based on edges. Then, GVF snake is used to extract the target contour, which is used to improve the color histogram. In the end, mean shift is improved combining with the Kalman filter or the particle filter based on the eolor histogram. The experimental results show the real-time performance in tracking non-rigid objects and the robustness to irregular movements and heavy occlusion.

关 键 词：非刚性目标跟踪 GVFSnake 均值漂移 卡尔曼滤波器 粒子滤波器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]