Stability Improvement in Tracking People by Introducing an Environment Model

We propose a method for tracking people in a cluttered indoor environment by integrating information from distributed sensors. Our method estimates the location and direction of a user's head by fusing multiple visual cues from multiple sensors based on particle filtering. Our method also utilizes a prior knowledge about the environment obtained from a laser range sensor for taking into account unevenly distributed probability of a user's position. The use of the environment model as well as integration of multiple visual cues from distributed sensors contributes to increasing tracking performance. We show the results of preliminary experiments that demonstrate the eectiveness of our proposed method.