A Method of Hand Contour Tracking based on GVF, Corner and Optical flow

Abstract

Since bare human hand is highly articulated and deformable, its tracking is more challenging than rigid object tracking. In this paper, an automatic approach is proposed to track hand contour on the basis of gradient vector flow (GVF Snake), corner detection and sparse optical flow. The skin-color model and motion information help to reduce the impact of local minimal under complex background. To lower the sensitivity of GVF Snake to initial contour, a curvature-difference-based corner detection is proposed to effectively detect enough corners on hand contour and sparse optical flow is applied to ensure a robust tracking of these corners even with violent gesture change. The experiments have shown a promising hand contour tracking using real image sequence under the usual lighting condition in lab.