Real-time road following in natural terrain

In the context of autonomous vehicle navigation, the use of colour content in images of off-road terrain within a tropical environment as a means of segmenting the scene is investigated. The operating environment is composed of dirt roads and tracks lined with vegetation with an irregular boundary between the two. Scene images are intelligently separated into driveable and non-driveable regions. HSI colour representation provides robustness to spectral colour variations and invariance to sudden changes in ambient light conditions. The approach to feature identification, classification and thresholding based on training set statistics is discussed. Finally results from the real world deployment on an autonomous vehicle are included