自适应窗口选取的Mean-Shift目标跟踪

当目标尺度发生变化时,传统Mean-Shift跟踪算法因跟踪窗口尺寸不变容易导致跟踪目标丢失,为解决此问题,本文提出一种带宽自适应算法对目标尺度变化进行检测,从而实现模板更新.该算法分别将模板图像与当前帧目标图像分割成等间隔半径的若干同心圆,通过计算模板图像与当前帧图像不同环层之间相似性度量,根据相应环层之间相似性度量关系确定当前帧模板带宽更新参量,最后利用kalman滤波完成模板尺度更新,从而实现目标稳定跟踪.实验证明,当目标尺度发生变化时,目标模板自动更新,能够实现目标稳定跟踪;相对传统Mean-Shift跟踪算法,目标跟踪可靠性能得到了提高.

Mean-Shift Object Tracking of Adaptive Tracking Window

To overcome the problem that traditional Mean-Shift based tracking method is easy to fail because of the rigid tracking window with the changing of object scales, a new algorithm to estimate the scale of the tracking window is proposed. It divides the template image and the current frame image into several concentric cirques. After calculating the comparability measurement of corresponding cirques between the template and the current image, it ascertains the bandwidth parameter of current frame. Finally, Kalman filter is adopt to finish the update of the template scale and the stable object tracking is realized. Experiments prove that the proposed algorithm is able to renew the template accurately when the scale of the object changes, and the tracking reliability is improved compared to the traditional Mean-Shift based tracking method.