Appearance Based Pose Estimation of AIBO’s

Robust techniques for pose estimation are essential in computer vision and robotic manipulation tasks. We present a novel approach for pose estimation based on angular color cooccurrence histograms and its application to robot players (Sony Aibo) in the Sony Four-Legged Robot League. The presented method is computationally efficient as no features have to be extracted from the image. The representation of objects is based on pixel cooccurrence histograms extracted from the color segmented image. In order to illustrate the principles of color cooccurrence histograms for pose estimation, the methodology is first explained for 2-DOF estimation for a rigid body object. The basic method is extended to pose estimation of the autonomous four-legged robots. Information on the pose of opponent robots allows a better prediction of the opponents behaviors and thereby opens an avenue for novel tactical options in robocup soccer.