Illumination recognition in field environments based on Gaussian Mixture Models

Terrain classification in field environment for mobile robot is affected by weather conditions among which illumination diversification plays a major role. With the changes in feature extraction method, classification result will vary significantly in the changing environment. This paper introduces a method of illumination recognition by analyzing the illumination distribution in visual space and using Gaussian Mixture Models to describe the illumination distribution. Illumination recognition provides a way to find suitable feature set for different illumination condition. The experiment result shows the illumination recognition with GMMs is accurate and efficient, and improves the performance of terrain classification.