Robust vision system with automatic filter calibration

In this paper we present an adaptive vision system for the AIBO robots which is able to dynamically change the calibration of the colour filters implemented in the robots. This system tries to be robust by using techniques for coping with noise and hard environment characteristics, specially changing lighting conditions. This system has to be run with a processor with low processing resources and therefore it has to be simple and efficient. The vision system uses the 8 colour filters that are hardware already implemented in the robots. These filters can be easily established and they are used to carry out the recognition of the relevant elements in the given scenario. The final goal of the proposed adaptive vision is to improve the performance of well-known self-localization algorithms.