An automated colour calibration system using multivariate Gaussian mixtures to segment HSI colour space

This paper presents a system for automating the time consuming task of manual colour calibration for a mobile robot. By converting a series of YUV images to HSI format and analysing histogram data it can be seen that there are distinct regions of colour space for each object colour class and that one dimension, hue, can be used to uniquely identify each colour class. Using an expectation maximisation (EM) algorithm to estimate the parameters of a Gaussian mixture model, it is proposed that the HSI colour space can be segmented and automatically labeled for the purpose of automatic colour calibration. This method is applied to a Aldebaran Nao robot vision system that uses a ‘soft’ colour classification method to classify non-unique colour space. By reducing the colour labeling dimension to one and implementing soft classification principles, a reliable automatic calibration system was achieved.