Hand Detection from Cluttered Images Based on a Hierarchical Strategy

Due to the variations of hand posture and the intricacy of environment/background, hand detection is a challenging task in human-computer and human-robot interactions. A hierarchical method is proposed in this paper to detect hand from images with cluttered background. In order to remove the most skin-like background of the image, upper body is detected from the image in the first hierarchy. Secondly, a novel approach, which combines samples threshold with experiential threshold, is proposed to detect skin/skin-like regions in images, and then skin regions are obtained according to the thresholds of area and length-width ratio of connected areas. At last, hand patches are determined by the hand model which is produced by support vector machine. The efficiency of this method is proved by corresponding experiments in hand detection in our dataset.