2003 Volume 16 Issue 2 Pages 77-84
In this paper, we propose a method of self-localization for mobile robots. The method is based upon model-fittng of floor region provided by the omni-directional camera mounted on the robot. In our method, even if a part of the floor region is hidden by obstacles, this can be compensated for by computing the convex hull of a boundary point set of the floor region in the omni-directional image. The geometric features of the detected floor region and the linearized least square method considering the properties of omni-directional imaging are employed to fit the known floor shape. In addition, we developed a self-localization method that allowed for better estimates of values by integrating omni-directional vision and dead reckoning with the Kalman filter than by individual methods. Finally, we verify the effectiveness of these methods through several experiments with a real robot according to the rule of the RoboCup Small Size League.