Skin Video Segmentation and Trajectory Recognition in Human–computer Interface Applications

In this paper we address two problems involved in video segmentation for human–computer interface applications. First we propose an adaptive method for hand skin segmentation to enhance the hand feature detector. A new analysis colour space is used in order to obtain a reliable adaptive skin detector. The colour space is chosen so that the image has a sparser representation than those using the classical colour spaces. Mean shift mode detection is used to segment the cluster corresponding to the skin. Secondly, a dynamic hand gesture segmentation approach is presented. We are concerned with locally linear hand gestures. Trajectory points are obtained from the hand feature detector. The trajectory is filtered in order to obtain a smooth continuous version. Combining Radon space analysis and a few static postures we generate an encoding sequence of the gesture that can be used further to attach an action to be performed by the computer.