A novel and practical approach towards color constancy for mobile robots using overlapping color space signatures

Abstract

Color constancy is the ability to correctly perceive an object's color regardless of illumination. Within the controlled, color-coded environments in which many robots operate (such as RoboCup), engineers have been able to avoid the color constancy problem by using straightforward mappings of pixel values to symbolic colors. However, for robots to perform color vision tasks under natural light the color constancy problem must be addressed. We have developed a color vision system which allows for the color space signatures of different symbolic colors to overlap. This raises the question: if a specific pixel value can be mapped to multiple symbolic colors, how does the robot determine which color is the "correct" one? Context plays an important role. We adopt a knowledge driven approach which allows the robot to reason about uncertain color values. The system is fully implemented on a Sony AIBO. © Springer-Verlag Berlin Heidelberg 2006.