A New Colour Space for Efficient and Robust Segmentation

Abstract

In many applications, real-time object identification and tracking is of paramount importance. Traditionally, such systems, employing real-time colour-based segmentation, are implemented in hardware to realise fast processing speed. Most of the inexpensive commodity frame grabber cards provide pixel colour information in RGB. However, using RGB colour space, it is difficult to separate the chrominance from luminance. For the classification algorithm to be robust in the face of variation of brightness of illumination, the HSI or YUV colour model is usually used. In this paper we evaluate the efficiency and reliability of using the classical methods to transform the RGB colour space to YUV colour space coupled with techniques to create and use efficient look-up-tables (LUTs) for fast colour classification of pixels. A new YUV colour space is proposed and its usefulness depicted by exhaustive data analysis using robot soccer system as the test platform.