Non-parametic model for robust road recognition

Road recognition is one of the key technologies in the vision-based intelligent navigation system. In this paper, we present a novel non-parametric estimation model and a robust approach for the unstructured road recognition. The model keeps a set of sample for both road region and off-road region, and then estimates the probability of a newly pixel based on color information. For improving the real time capability and ruling out the interferences caused by variances of illumination and shadows, the image is divided into several small blocks, and a segment method is used to extract the lane boundaries from the mixed block areas. Finally, the boundaries of the lanes are fitted by the B-spline curve in which the best control points are searched by the least square method. Both field tests and simulation show that the proposed algorithm is effective and robust.