自适应窗宽的均值迁移图像跟踪算法

针对传统均值迁移跟踪算法中核函数窗宽固定使之无法满足图像中运动目标尺寸变化需要的问题，在分析目标特征零阶矩特点基础上，以其变化比作为观测量，以目标面积的变化比作为状态量，利用卡尔曼滤波器对未来帧中目标面积的变化进行预测进而获得同目标尺度变化相适应的核函数窗宽。该算法通过窗宽自适应变化，提高了跟踪精度，增强了跟踪稳定性，同时仍保证了跟踪的实时性。实验结果证明了该方法的有效性。

Mean-shift image tracking algorithm with the adaptive bandwidth

The classic tracking algorithm based Mean-shift uses fixed kemel-bandwidth, which limits the application when the object scale exceeds the size of the tracking window. Based on analyzing zeroth moment of the target feature, Kalman filter is introduced in the algorithm to forecast the change of object area for adaptive bandwidth, which the zeroth moment change of the target feature is regarded as the observation and the area change of target is regarded as the state. Thus the appropriate kernel-bandwidth being sea- soned with the change of object scale is adaptively acquired. The algorithm improves the traeking precision and stability with assuring real-time. The algorithm is proved effective by experimentation results.