Research On Hand Detection And Tracking

Human-Computer interfaces (HCI) have evolved from text-based interfaces through2Dgraphical-based interfaces, multimedia-supported interfaces, to full fledged multi-participantVirtual Environment (VE) systems. While providing a new sophisticated paradigm for humancommunication, interaction, learning and training, VE systems also provide new challengessince they include many new types of representation and interaction. The traditionaltwo-dimensional HCI devices such as keyboards and mice are not enough for the latest VEapplication. Devices that sense body position and hand gesture, speech and sound, facialexpression, and other aspects of human behavior or state can be used so that thecommunication between the human and the VE can be more natural and powerful.Gesture is a powerful human-to-machine communication method. Two primary problemsof gesture recognition are hand detection and hand tracking. The tasks of hand detection andtracking are challenging, because the hand is a non-rigid object. Considering the hand poseand each finger joint, the human hand motion has roughly with27degrees of freedom.Several hand detection and tracking systems had been proposed. The first generationapproaches require glove-based devices to help recognize the hand. However, the gloves andtheir attached wires are still quite cumbersome and awkward for users. Moreover, the cost ofthe glove is often too expensive for regular users. The second generation approaches use skincolor or shape feature. However, those methods are lack of robustness when dealing withdynamic environments and various kinds of lighting. The third generation approaches arebased on a cascade architecture using boosting algorithm, which was first introduced by Violaand Jones for face detection and tracking problems. That approach allows robust and fastdetection of hands.In this paper my research is focused on the hand detection and hand tracking. First, I willintroduce the cascade architecture for object detection. Then I will introduce two new featuresfor hand detection. Finally, I introduce a hand tracking method based the CONDENSATIONand feature detection algorithm. Experiment result shows our method can get a real-time androbust hand tracking performance.