Illumination independent object recognition

Abstract

Object recognition under uncontrolled illumination conditions remains one of hardest problems in machine vision. Under known lighting parameters, it is a simple task to calculate a transformation that maps sensed values to the expected colors in objects (and minimize the problems of reflections and/or texture). However, RoboCup aims to develop vision systems for natural lighting conditions in which the conditions are not only unknown but also dynamic. This makes fixed color-based image segmentation infeasible. We present a method for color determination under varying illumination conditions that succeeds in tracking the objects of interest in the RoboCup legged league.