Human-of-interest tracking system for natural interaction

There have been considerable endeavors for robustly tracking human (his/her whole body's or body parts') gestures for achieving natural human-computer interaction but it is still a challenging problem in vision-based research area. In particular, tracking selectively one's-of-interest (or his/her body parts') gestures, which is necessary in multi-user applications, is more difficult and has been less developed. In this paper, we propose a simple yet effective method for human-of-interest tracking, which combines a background subtraction method using difference keying, a skin region detection method in Y CbCr color space, a region-of-interest detection method using the de pthinformation from infrared images, and a region classification method based on a set of Haar-like features and Adaboost learning algorithm. Based on the proposed method, we built a proof-of-concept system. The system proved to be able to provide natural frontalview interaction with one of interest in real-time and robustly in a cluttered indoor and multi-user environments. We expect that our interaction system can provide a practical solution for natural interaction with digital in formation d isplay systems and interactive game or education systems.