Hy-Seg: A Hybrid Method for Ground Segmentation Using Point Clouds

Abstract

Real-time and accurate ground segmentation is a crucial technology for intelligent vehicles. On the one hand, it obtains drivable area information for vehicles, which is of great significance for subsequent navigation and control. On the other hand, it is beneficial for improving the performance of object detection and segmentation. In this paper, Hy-Seg is proposed, which is a hybrid method and achieves accurate and fast ground segmentation. First, to improve the efficiency of the algorithm, the 3D point cloud generated by LIDAR is represented by a polar grid map. Then, candidate ground points in each grid are generated by a gradient-based method, reducing most of the false-positive results. Polynomial fitting based on random sample consensus is used to model the ground of each segment correctly. Finally, the original 3D point cloud is segmented using the fitted model. Experiments on SemanticKITTI show that the proposed method can not only achieve an accurate segmentation result but also run at a frequency of 53 Hz, which meets the requirements of intelligent vehicles.