In mobile robotics, it is necessary to have a robust and efficient way of describing the visual stream provided by a vision system to be used afterwards in tasks such as object recognition. Color histograms are a useful tool to capture and represent color properties of sets of images taken from a certain position. Since those images were obtained at different time and light conditions, their appearance have greatly changed, reducing the performance of the color descriptor. In this work, we develop a color constancy algorithm that copes with the color variation among sets of images taken from nearly the same place. We show that the performance of the color histogram descriptor rises after color constancy, becoming a more robust and useful color descriptor. In the results section, we support that claim with several sets of images of scenes belonging to different positions.

This work was supported by the project 'Integration of robust perception, learning, and navigation systems in mobile robotics' (J-0929).

Trabajo presentado al 9th CIARP celebrado en Puebla (Mexico) del 26 al 29 de octubre de 2004.

Peer Reviewed