Finger-fist detection in first-person view based on monocular vision using Haar-like features

This paper introduces a new idea for interaction between human and wearable device which is using finger-fist posture to be the detecting and tracking target. We built the detector with cascade classier using Haar-like features and the AdaBoost learning algorithm. The detector for the posture shows good tolerance for out-of-plane rotation and robustness against lighting variance and cluster background. With excellent real-time performance and high recognition accuracy, the detection can be acted as a tracker to track the path of fist in the first-person view.