Hand tracking in unconstrained environments remains an extremely challenging problem due to several factors, such as background clutter, deformation, and motion blur. In this paper, we combine motion, color, and Haar-like features to construct a real-time and robust hand tracking system. Haar-like features successfully defeat moving skin-colored backgrounds, although they are unstable for the whole situation. Three weak trackers are built using each kind of feature and integrated in a boosted cascade. If one stage makes sure of the object position, no other stages is carried out. Otherwise it provides its own point of view to guide the next stage. We realize the proposed approach and demonstrate it on several challenging sequences.%由于复杂背景、形变以及运动造成的模糊等因素,导致在自然场景下的人手跟踪仍然是一个具有挑战性的问题.本文中,结合运动、颜色和Haar-like特征来构造一个具有鲁棒性的实时人手检测算法.尽管不能运用于所有的情形,但Haar-like特征成功地去除了类似肤色的运动背景区域.利用三个特征构造三个弱分类器,然后将其结合成一个强分类器.如果一个分类器已经确定了人手的位置,其他分类器将不会执行,否则将会为下一个分类器提供一个可能的区域.文中实现了提出的算法,并且在几个具有挑战性的视频序列上进行了实验.
展开▼
