Abstract: Under-segmentation and over-segmentation exist in the process of road segmentation for the environment system of intelligent vehicle. To solve such problem, a road segmentation method based on loopy belief propagation algorithm is proposed. First, multiple label Markov Random Field is applied to build grid map in the polar coordinate system, and several measurement cost function models are constructed based on the point distribution in each grid cell. Next, the local smoothness and slop hypothesis are integrated to establish smoothness cost function model and ensure the discontinuity of the ground height among the grids. Finally, the loopy belief propagation algorithm is employed to transmit the message repeatedly, and the road height of each grid is estimated with the largest belief to segment the road points. Experiment results show that the proposed method has better road segmentation performance than other methods, especially for rough or sloped road conditions.

Key words: vehicle engineering, road segmentation, loopy belief propagation, Markov random field, cost function, point cloud distribution