Calling Motion and Natural Hand Detection for Gesture Recognition

摘要

Most existing gesture detection and tracking algorithms construct fixed postures that make search regions narrower. Under robot circumstances, however, it is unreasonable to makes users stick to a certain pose. In this paper we suggest a particular motion that help to detect and tracking hands easier under robot circumstances. The requirements of such motion are mainly three; first, the motion can be detected with light computation because the motion is executed in the robot almost all the time. Second, the motion is easy to perform. That is, the pose should be one of the gestures naturally taken and used. Lastly, the pose should be guaranteed to detect with few efforts anywhere and anytime. We propose calling behavior as the motion for naturally detecting hands and tracking. Because calling motion contributes to trigger gesture recognition and other applications of human robot interaction as well as it support to find skin colors adapted to the situation where calling motion is identified. Finally, we experiment on the validity of calling motion with hand detection and tracking after identifying calling motion