A NEURAL APPROACH TO REAL TIME COLOUR RECOGNITION IN THE ROBOT SOCCER DOMAIN

Abstract — This paper addresses the problem of colour recognition in real time computer vision systems applied to the robot soccer domain. The developed work presents a novel combination between our current model-based computer vision system and neural networks to classify colours and identity the robots on an image. This combination presents an alternative to the considerably high computational cost of the neural network, which reduces the demanding and time consuming task of calibrating the system before a match. The results prove that this new approach copes with luminance variations across the playing field and enhances the accuracy of the overall system