窗宽自适应Mean-Shift跟踪算法

摘要

针对固定窗宽Mean-Shift算法在目标运动速度过快或尺度发生明显变化时可能导致跟踪失败的问题,提出一种窗宽自适应的Mean-Shift跟踪算法.该方法基于均值漂移矢量预测跟踪窗口中心位置,同时自动调整跟踪窗口大小,保证目标始终处于跟踪窗口内部,使算法得以准确定位目标;在确定空间位置后,利用基于Bhattacharyya系数的二分法自动选取窗口缩放比例,得到与目标尺度一致的跟踪窗口.实验结果证明,该方法能很好地定位目标的空间位置和尺度.%Mean-Shift algorithm with fixed bandwidth often fails in tracking an object that moves too fast or owns a dramatic change in scale. To solve the problem, a new Mean-Shift tracking algorithm based on adaptive bandwidth was proposed. Mean-Shift vector was used to predict the center position and automatically modulate the size of tracking window that fixed the object inside the window and gained accurate object position. After confirming the position, a Bhattacharyya coefficient based dichotomy was adopted to select the pantograph ratio automatically, and a tracking window adapted to the scale of object was obtained. The experimental results prove the algorithm' s capability in locating object' s position and scale.