一种特征融合的带宽自适应Mean Shift跟踪算法

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。

A Feature Fusion and Automatic Bandwidth Selection Mean Shift Target Tracking Algorithm

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. We propose an improved feature fusion and automatic bandwidth selection algorithm in this paper. For traditional color histogram is vulnerable to the background region of object area, edge histogram is integrated and kernel bandwidth is updated adaptively in this paper. In addition, this paper has done some further research on updating target model and gives some comparisons. Results show that the algorithm has a stable performance on object tracking in scenarios color characteristics of object and background are similar and scale of object changes.