A method for following unmarked roads

Road-following for ground vehicles has been widely studied and the solutions normally tracks line markings or a salient road edge. Most of the presented approaches are thus based on geometric or spectral models and does a fit of image data to the model to ensure robustness. For operation on unmarked roads under highly varying weather and light conditions a strategy is presented which is based on a probabilistic generative model. The method is evaluated on a number of test scenes and compared to well known methods from the literature.