Abstract: In order to improve the adaptability of road model for complex environments, achieve a balance between efficiency and robustness in urban semi-structured road environment, a novel road detection algorithm based on uncertain Bezier deformable template is improved. It adaptively sets two-tier region of interest containing static and dynamic area for IMP according to camera parameters and vehicle status and proposed Gaussian mixture directional filter for pretreatment. It constructs uncertain road deformable template using Bezier curve, and convert the road recognition problem to template parameter hypothesis testing problem. A modified RANSAC algorithm is also wsed to solve the uncertain deformable template parameters. In order to improve the speed of solving, a level search optimization algorithm which using a combination of coarse search and fine search method is proposed. The test results show that the method can rapidly and accurately extract the lane and have good immune function for typical roads interference in the urban road environment.

Key words: Gaussian mixture directional filter, Lane detection, Modified RANSAC algorithm, Uncertain Bezier template