Robust Color Classification for Robot Soccer

This paper presents an adaptive approach to improve the reliability and performance of color classification. Therefore, we transform the camera-data from YUV into a novel chrominance space. Thereby, an optimized transformation function is given by an evolutionary algorithm. The novel idea is, not to adapt the thresholds that define a specific color region, but to evolve an optimal chrominance space transformation. In the novel chrominance space, the color regions are located in easy-toseparate subspaces which reduces the algorithmic complexity of color segmentation and improves classification accuracy significantly.