Abstract
One of the most important aspects in mobile robots concerns vision based decision-making systems, where color constancy is a major problem for robots, since they often use the color property to recognize their environments. Animals can recognize the color and shape of objects despite large changes in lighting conditions in outdoor environments, for example. Biomimetic software and hardware have attracted much attention because of the possibility of realizing flexible and adaptive systems as in animals. We have been working on color constancy vision algorithms using a bio-inspired information processing technique. In this paper, we evaluate the performance of color recognition using bio-inspired information processing algorithms, namely the Self-Organizing Map (SOM), modular network SOM (mnSOM), and Neural Gas (NG), and discuss the dynamical experimental results under various lighting conditions.
