基于最小软阈值二乘的目标跟踪方法能够较好地处理视频的外观变化和异常值,但当目标子空间受到姿态变化或遮挡等干扰时,跟踪器的稳健性较差。针对这一问题,在贝叶斯引理框架下,提出一种组合最小软阈值二乘和压缩Haar-like特征匹配的在线目标跟踪算法。该算法针对最小软阈值二乘跟踪器采用定量遮挡率来评判其观测样本受离群子干扰程度,并在跟踪器单帧匹配响应过低时,利用压缩特征匹配对观测目标进行二次筛选。同时,通过观测置信度减少无关样本的数量,降低计算复杂度。实验结果表明,本文提出的算法能够取得更加优异的跟踪结果。

The object tracking method based on least soft-threshold squares deals with the appearance change and outlier of video well. However, when the object subspace is influenced by interference such as posture change or occlusion, the tracking robustness is not completely effective. To solve this problem, this study proposes an online object tracking algorithm which combines least soft-threshold squares with compressed Haar-like feature matching in the framework of Bayes lemma. First, we employ the quantitative occlusion for the least soft-threshold squares based tracker to measure the extent of interference of outlier of observed samples. Then, we sieve the observed object again with the compressed Haar-like feature matching when the single-frame matching response of the tracker is very low. Meanwhile, by reducing the number of independent observed samples through the observed confidence coefficient, the computation complexity can be reduced. The experimental results show that the proposed method can be more effective than other methods.