Visual perception tools for natural interaction - a gaze capture and tracking system

In this paper we present a vision system with the aim of building a perceptual tool in order to make machines aware of humans allowing natural interaction in human-machine communication. The system must satisfy interaction constraints and be non-intrusive. Such a real-time camera-based system is designed for gaze tracking from an image sequence of a human in front of a machine. Therefore we use a CCD camera, placed between the keyboard and the screen. The system detects the user's presence, locates and then tracks the face, nose and both eyes. These operations are performed by combining image processing techniques and pattern recognition methods. The tracking process is based on a prediction-verification strategy using dynamic information of detection. The system continuously adapts its recognition parameters to take into account environment variations. Some results of recognition scores are also presented.