Abstract: In order to track moving targets of video exactly, a novel tracking scheme is proposed using particle filter based on improved color histogram features and Speeded Up Robust Features（SURF）. SURF algorithm is applied to extracting features. The features are tracked stably by the hierarchically iterative matching of the Kanade-Lucas-Tomasi（KLT） matching algorithm. Based on?fusion mode of multiplicative weights, the SURF and improved saliency weighted color histogram are used to the joint observation probability. Meanwhile, for avoiding the drop of SURF matching numbers and the instability in tracking process, a SURF feature-point-set update method is presented. Through using multi-group experiments, the results demonstrate that this approach has great robustness for either sunlight or block and has high accuracy in tracking.

Key words: target tracking, visual saliency, color histogram, Speeded Up Robust Features（SURF）, particle filter, KLT algorithm

摘要： 针对视频中运动目标的准确跟踪问题，提出了一种改进的颜色直方图特征和SURF特征的粒子滤波跟踪算法。采用SURF算法提取特征点，利用分层迭代的KLT算法对特征点进行稳定跟踪。将SURF特征与改进的视觉显著性颜色特征进行乘性融合，作为粒子滤波的观测概率。针对跟踪过程中SURF匹配数下降和不稳定的现象，设计了SURF特征模板集的更新策略。与传统特征的跟踪进行多组对比实验，其结果证明了该方法对光照和遮挡具有很好的鲁棒性，对目标跟踪的准确率更高。

关键词: 目标跟踪, 视觉显著性, 颜色直方图, 加速鲁棒特征（SURF）, 粒子滤波, KLT算法