基于前景概率函数的目标跟踪

针对不规则目标跟踪中初始窗口内包含背景像素导致特征模板不准确的问题,提出前景概率函数以及基于前景概率函数的目标跟踪算法.首先根据目标所在区域与背景区域的颜色分布建立前景概率函数,并以此计算目标区域中像素的前景概率,削弱背景像素的干扰,得到更准确的目标特征模板.将目标区域像素的前景概率引入均值迁移跟踪框架中,实现目标的迭代定位;在跟踪收敛后重新计算收敛区域中的前景概率分布,根据其反向投影图的尺度变化调整跟踪窗宽;最后利用Bhattacharyya相关系数对目标特征模板进行自适应更新.实验表明,该算法能够有效抑制背景像素的干扰,在目标尺度变化时能够准确调整跟踪窗宽,减少迭代次数,满足实时跟踪的需要.在复杂背景中跟踪性能也始终优于传统的均值迁移跟踪算法.

Object Tracking Based on Foreground Probability Function

In irregular object tracking, the precise model is not available because of the pixels belong to background in the tracking box. A novel tracking method based on foreground probability function is proposed to solve the problem above. Firstly, the foreground probability function is constructed to calculate the foreground probability of each pixel in tracking box, according to the color distribution of the foreground and the background region nearby. A precise model is obtained because the background pixels are excluded. Secondly, the foreground probability map is introduced to the mean shift framework to track with the object. After a tracking unit is completed, the scale of the tracking box is adjusted according to the new foreground probability map. Finally, the Bhattacharyya coefficient of the adjacent frames is used to modify the object model adaptively. Experimental results show that the foreground probability function is effective to suppress the compactness of the background pixels in the tracking box. Compared with the conventional tracker, the proposed tracker is robust to the scale change, and needs less iteration, which meets real-time tracking condition. In the tracking test of video sequence with complex background, the new tracker works better then conventional mean shift algorithm.