一种特征融合的带宽自适应Mean Shift跟踪算法  

Improved Target Tracking Method Based on Fearture Fusion and Automatic Bandwidth Selection Mean Shift Algorithm

作　　者：张良[1] 郑世宝[1] 杨华[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》2012年第5期118-121,共4页Video Engineering

基　　金：上海市科委重点攻关定向项目(10231204002)

摘　　要：传统Mean Shift跟踪算法在目标颜色特征和背景颜色特征相近、尺度变化等情景下效果不理想。提出了一种特征融合且核函数带宽自适应的改进跟踪算法,针对颜色直方图容易受背景区域影响,融合了边缘直方图,并对核函数带宽进行自适应更新。另外,对目标模型的更新进行了相关探索并给出比较结果。结果表明,该算法可以对目标实现更稳定的跟踪,对目标颜色和背景颜色相近、尺度变化等场景具有很好的适应性。The effect of traditional Mean Shift tracking algorithm is not satisfactory in scenarios when color characteristics of object and background are similar and scale of object changes. An improved automatic bandwidth selection algorithm is proposed in this paper. And for that traditional color histo- gram is vulnerable to the background region of object area, edge histogram is integrated and kernel bandwidth is updated adaptively. In addition, some further research on updating target model is given, and some comparisons are made. Results show that the algorithm tracks object more accurately in sce- narios when color characteristics of object and background are similar and scale of obiect chan^es.

关 键 词：目标跟踪 Mean SHIFT 颜色直方图 边缘直方图 核函数带宽 

分 类 号：TN919.8[电子电信—通信与信息系统]