Object recognition under various lighting conditions

Abstract

This paper describes object recognition under various lighting conditions for a robot which finds a user-specified object and brings it to the user. The system first constructs object models under a known lighting condition. Because the lighting condition in model construction is different from that in object recognition, the system needs to transform the model's color for the current lighting condition. The system estimates color transformation by using only one observed color of a reference object. Last the system tries to recognize a user-specified object based on the estimated lighting condition. When the system fails in recognition, it tries again by using the user's advice. Experiments using real-world refrigerator scenes are shown. © Springer-Verlag; 2003.