Adaptive and real-time unpaved road segmentation using color histograms and RANSAC

The determination of the position of the road that a vehicle is following is a topic of high interest in research applications related to driverless cars and driving assistance tools. The complexity of this task increases in natural terrains with unpaved lanes and unstructured objects such as trees, stones, and other natural objects. In this paper, a novel, adaptive and real-time visual road segmentation method for off-road autonomous driving is proposed. The method is based on the use of adaptive color histograms, however, its main novelties are the use of the RANSAC algorithm in order to determine the road borders and the vanishing point, and a set of heuristic rules that allow checking the spatiotemporal consistency of the road determination. Both ideas allow obtaining, at the same time, high robustness and real-time operation. The method is validated using real-world video sequences obtained in off-road driving.