基于自适应活动轮廓模型的实时手势跟踪

Real Time Hand Tracking Based on Adaptive Active Contour Model

在基于视觉的手势分析与识别中,一个关键环节是手势跟踪。本文提出了基于颜色信息的自适应活动轮廓模型,并与均值漂移算法相互融合,实现图像序列的实时手势跟踪。跟踪算法分为两步进行,首先应用均值漂移算法实现手部区域的定位,然后基于自适应活动轮廓模型提取手部轮廓。在跟踪过程中,轮廓提取为下一帧的区域定位更新搜索窗口,提高了搜索效率,使目标跟踪达到实时性要求。同时,本文根据跟踪区域模板与目标模板的相似性度量Bhattacaryya系数给出了在跟踪目标被遮挡时的处理方法,有效地解决了这一难题。实验结果证明了在无遮挡和遮挡两种情况下算法均能实现准确、实时的手势跟踪。

Hand tracking is an essential step for vision based gesture analysis and recognition. This paper presents an adaptive active contour model using color information, which is connected with mean shift algorithm to implement real time hand tracking in sequences. The proposed method consists of two steps： hand location using mean shift and hand extraction based on adaptive active contour model. In the process of tracking shape extraction alters the search window for hand location of next frame, which improves searching effective and makes tracking real-time. At the same time, this paper gives tracking methods in terms of the similarity measure of candidate modal and object modal i.e. Bhattacaryya coefficient while the object is occluded. Experimental results show that accurate and real-time tracking is achieved using the proposed algorithm either on the occasion of occlusion or not.