Evaluation of color space for segmentation in robot soccer

Abstract

Color segmentation plays an important role as a primary process in the vision system of the robot soccer, since color is an effective and robust visual cue in the complex environment. However, color segmentation suffers natural variations in light and shadows as well as the limitation of calculation due to the real-time robot system. A color transformation is assumed to increase separability among different color class labels, for color segmentation approaches are highly dependent on the robustness of the used color spaces to these variations. In this contribution, measures based on geometric distance, histogram comparison and information theory are employed to evaluate a color space. Ten color spaces are compared by evaluation metrics for robot soccer vision system in the RoboCup Standard Platform League (SPL). The experimental results reveal that the optimal performance was obtained by the transforming the pixels to the CIE-L*a*b* and RGB color spaces. Moreover, the linear transform color spaces have better performance than perceptually color spaces according to the metrics for RoboCup SPL.