基于统计分类的多行人跟踪算法

为了使视频监控系统具备多行人自动跟踪能力,提出了一种基于统计分类的多行人跟踪算法。首先采用了背景差和Canny算子准确提取出运动区域,然后利用特征匹配实现帧间目标的对应,最后提出一种利用人体表面颜色空间分布信息的统计分类方法,完成合并或遮挡时多个行人的分离和单独跟踪。实验表明,在相互遮挡时,该算法也能有效地分割和跟踪多个行人,适于智能视觉监控应用。

Multiple People Tracking Based on Statistical Classification

In order to endow visual surveillance systems with the ability of multiple people tracking,a multiple people tracking algorithm based on statistical classification is proposed.The algorithm adopts background subtraction and Canny operator to extract the moving regions.Then,features-matching approach is used to find the temporal correspondences of segmented objects between successive frames.Finally,a statistical classification method which exploits color spatial distribution of the human body is proposed to separate occluded people and track each individual.Experimental results show that this algorithm can effectively segment and track multiple people successfully through occlusion and is suitable for intelligent surveillance.