针对不规则目标和背景区分度低的目标在跟踪过程中尺度和旋转自适应的问题,提出了基于前景概率密度函数和椭圆拟合的均值平移目标跟踪算法。首先依据初始化的目标窗口及其周围的背景区域建立前景概率密度函数,抑制跟踪窗口内的背景像素,建立可靠的目标模板；然后将前景概率密度函数引入均值平移的迭代过程,实现跟踪窗口的快速定位；最后计算跟踪窗口内各点的前景概率密度函数,反向投影为前景概率分布图,通过对该图的边缘提取和椭圆拟合,获得当前目标的位置、尺度和旋转信息,并将其用于对下一帧跟踪窗口的初始化。实验结果表明,该方法克服了背景干扰对跟踪窗初始定位和目标信息更新的影响,能够实现对目标尺度和旋转的自适应,具有跟踪稳定和实时性高的特点。

Aiming at solving the problem of scale and orientation adaption in irregular target and low discriminating target tracking,a mean shift tracking algorithm combining foreground probability function and ellipse fit was proposed. Firstly,a foreground probability function was established to weak the background pixels in the window and to obtain an accurate target template,according to the initial target window and the surrounding background. Secondly,the foreground probability function was introduced into the iteration of mean shift,to achieve the position of the target window quickly. Lastly,the foreground probability function of each pixel in the final target window was recomputed and then projected to the foreground probability image,in which the edge detection and ellipse fit were orderly applied to update the ellipse of target with current optimal position,scale and orientation. The present tracking result was used for initialization of the next frame. The experimental results indicate that the proposed algorithm overcomes the background interference on the coarse position of target window and information update of target,can achieve stable and realtime tracking with adaption to scale and orientation.