2008 Volume 21 Issue 1 Pages 1-9
This paper proposes an accurate self-localization method for a mobile robot. The position of the robot mounted a pan-tilt camera is estimated from the images taken by the camera. In conventional self-localization methods using a stable camera, the detection error of landmark positions on the image was a big problem. Then, we focus on the landmark trajectory on the sequential images which are taken by rotating the pan-tilt camera. By fitting a curve to the trajectory points, the more accurate positions of the landmarks can be obtained than using the multiple images taken by the stable camera. In addition, there is a problem of the landmark displacement on the image due to the tilts of the camera rotation axes. In order to solve this problem, the method for calibrating the tilts using the landmark trajectories is proposed. Our proposed methods are validated for the accurate self-localization in the experiments using the artificial data and the real images taken by the pan-tilt camera.