Color vision for RoboCup with fast lookup tables

Abstract

The small-size league of the RoboCup robot soccer competition requires a machine vision system capable of tracking and locating up to 11 objects in high-resolution color images at video frame rates. The problem is eased by the restriction that all objects are marked with solid color patches. This paper describes a high speed color vision processing system with distortion correction used by Team Canuck of the University of Alberta to compete in RoboCup 2002. We specifically focus on how color classification is performed adaptively and efficiently for real-time robot soccer, using fast look-up tables defined in color subspaces.