视频图像序列中人体目标的提取与跟踪

针对人体目标运动的非刚体性而难以跟踪问题,提出了一种基于视频图像的人体跟踪方法。首先获取人体目标,选取目标若干子区域作为跟踪模板,用Kalman滤波和最小差值法作为相似性测度的相关跟踪对各子区域进行精确定位,最后将各个子区域跟踪结果进行位置融合得到最终的人体目标位置。实验表明,该方法对人体目标细节性形变和大的形变均能进行有效跟踪。

Extracting and Tracking the Human Object in Video Sequences

Human object tracking is a difficult problem because of its non-rigid nature.In order to solve this problem,a new method was presented.In this method,human object was firstly extracted and then several sub-regions was selected as tracking templates.In the following,Kalman filter combined with correlation tracking which used minimum difference method as the similarity measurement were used to locate these sub-regions accurately.Finally all these results were fused to get human object＇s ultimate location.Experiments showed that both little and large deformation of the human object could be effectively tracked.