Real Time Hand Tracking System Using Predictive Eigenhand Tracker

In this paper, we present a real time vision based hand tracking system by combining predictive framework and appearance model. Due to the nature of hand motion which is flexible, erratic and easily varies in its appearance, the hand tracking from a single camera remains a complex problem. Here, we present a simple and efficient method to overcome such difficulties using the integration of Adaptive Kalman Filter (AKF) and Eigenhand method. After the hand state is quickly estimated from the AKF prediction, appearance model is employed to improve the earlier estimation. The appearance model is constructed based on a low dimensional eigenspace representation; the so called Eigenhand. During the tracking, the eigenspace constantly learns and adapts to reflect the appearance changes of the hand image. The experimental results demonstrate the effectiveness of the proposed tracking algorithm in indoor and outdoor environments where the target objects undergo large pose changes, lighting variation and partial occlusion. We achieve an average detection rate above 97% at the speed of 35fps.