Object recognition using lighting condition database based on long-time observation in virtual environment

Object recognition by general real robots needs to compare a captured image to reference images. If a lighting condition is different from the one in the database, the robot should measure the parameters of lighting condition or use huge image database which covers so many lighting condition in order to recognize the objects. However it is difficult for real robots. In this study, we propose a novel approach to estimate the lighting conditions of objects by using lighting condition image database based on long-time observation in virtual reality. We also propose an approach to predict the lighting condition of a new object image captured in unknown lighting conditions by using lighting condition filters calculated from lighting condition image database. SIGVerse was used as simulation platform to effectively develop and evaluate our proposed approach. Proposed approach was also evaluated in actual lighting conditions and environments.