Bridging gap between the simulation and robotics with a global vision system

A wide gap lies between the simulation league and real robot leagues of RoboCup, in particular in applications. In the simulation league, the soccer server maintains all data on the pitch and feeds such data to applications via networks. Therefore, several applications such as 2D- and 3D-viewers, commentary systems have been developed. In this paper, we propose that using a global vision system can bridge the gap of applications. A global vision system, which is usually used in the small size league, is extended to serve as a data feeder. Its main issues include missing object identification (due to occlusion and feature extraction failure), errors in player identification (due to mis-assignment of objects among continuous frames), detection of stoppage, and elimination of unnecessary scenes. To cope with these issues, the proposed system uses estimation and scene analysis. We also present an extension of the soccer server protocols to feed state specific to the small size league. The resulting system succeeded in feeding data of the final game of RoboCup'99 to a 2D-viewer and a commentary system.