Lidar based perceptual system is an important aspect of autonomous driving. In this paper we focus on ground segmentation and road boundary extraction in point cloud. In order to solve the problem that most of the current ground segmentation algorithms can not make full use of the data information, which leads to poor segmentation results on complex roads (such as slopes and undulating roads), this paper presents a two-stage ground segmentation algorithm that combines the Cloth Simulation Filtering (CSF) algorithm and the 2.5D grid map method. In the first stage, CSF is applied to filter out most ground points. Then we project the non-ground points obtained in the first stage to a grid map and judge the attribute of each grid in the map for further segmentation refinement. After ground segmentation, road boundary extraction also plays an indispensable role to obtain drivable area and reduce search space. However, most of road boundary extraction methods focus on single frame data and discard useful historical information. In this paper, the road boundary seed points of the current frame are accumulated from previous multiframe point clouds. After that, the curbs are fitted via the parabola model and RANSAC algorithm. Extensive simulations in Prescan show that the proposed ground segmentation algorithm can adapt to various complex road conditions. Field tests were also conducted to demonstrate the effectiveness of the proposed segmentation method. Our experimental results of road boundary extraction also show that after multiframe accumulation, more detailed information of road boundary can be obtained, resulting in more accurate detection results. In straight part the extraction accuracy of road boundary is more than 96%; in the curve part, the extraction accuracy of road boundary is not as stable as the straight part, but the lowest extraction accuracy is 89.33%, and the average extraction accuracy is 95.64%. Finally, we analyze the influence of localization and attitude angle error on road boundary extraction.

Mr. Congcong Li, Tongji University, CHINA Prof. Dr. Lijun Zhang, Tongji University, CHINA Prof. Dejian Meng, Tongji University, CHINA