Histogram-Based Visual Object Recognition for the 2007 Four-Legged RoboCup League

Abstract

RoboCup is an annual international robotic soccer competition. One of its most popular divisions is the Four-Legged League, whereby each team consists of four fully autonomous Sony AIBO robots and researchers focus solely on software development over a standard robotic platform. To compete successfully each robot needs to address a variety of problems: visual object recognition, legged locomotion, localization, and team coordination. This paper focuses on the problem of visual object recognition and suggests a uniform approach for recognizing the key objects in the RoboCup 2007 field: the two goals, the two beacons (colored landmarks), and the ball. The proposed method processes the color-segmented camera image and delivers the type, as well as an estimate of the distance and the angle with respect to the robot, of each recognized object in the current field of view. Our approach is based on a number of procedures which are used uniformly for all three types of objects: horizontal and vertical scanning of the image, identification of large colored areas through a finite state machine, clustering of colored areas through histograms, formation of a bounding box indicating possible presence of an object, and customized filtering for removing implausible indications. Our approach is compared against the approaches of two RoboCup teams (German Team 2004 and SPQR-Legged 2006) and is shown to be equally good or better in many cases. The proposed approach has been used successfully by Team Kouretes of the Technical University of Crete, Greece during various RoboCup events.

This project was partially supported by the European Marie-Curie International Reintegration Grant MCIRG-CT-2006-044980 awarded to Michail G. Lagoudakis.