Multi Gesture Recognition: A Tracking Learning Detection Approach

Abstract

Many real world activities involve the interactions of multiple gestures, e.g., jogging on the playground with both legs and waving hands, saying good bye with shaking both hands, etc. However, current vision based gesture recognition algorithms assume there is only single gesture in the scenario. Some existing multiple gesture recognition detection systems require the aid of particular devices such as multi-touch pad, infrared sensors, gyroscope sensors etc. In this paper, we proposed a new tracking learning detection framework for recognizing the multiple gestures in the video stream. The framework is based on tracking learning detection (TLD) [1] approach, which integrates the short-term gesture tracker and online learned gesture detector. With the collaboration of tracker, detector and online learning algorithm in TLD, it can be successfully adapted to vision based multi-gesture recognition. Experiments show that our framework outperforms detection based methods with vision based multi gesture recognition.