Human hand tracking from binocular image sequences

Sensing of human hand motion is very important for a variety of applications, such as computer animation and athletic performance measurement. Tracking a hand is difficult because the hand has high degree of freedom articulated mechanisms. This paper presents a 3D model-based hand tracking method which is robust to occlusions and local minima. Tracking is performed by minimizing the estimation error of an optical flow and maximizing the overlap between a projected model and a silhouette image. The authors employ stochastic optimization to solve them, which are generally difficult. They present experimental results on tracking from synthetic and real image sequences