We propose a ground segmentation algorithm for various urban environments to overcome the problems of under-segmentation and over-segmentation on sloped roads, obstacles and ground junctions. First, the point cloud is linearly arranged based on the horizontal angular resolution of the lidar. Then, the ratio of the distance from front and rear points to the lidar is used to remove abnormal noise, and the height threshold is adaptively adjusted by using the distance from each point to the lidar and slope value. Finally, ground segmentation is performed using the adjusted global and local height thresholds. Through experimental analysis of three different types of urban roads, it is verified that the proposed algorithm can effectively distinguish the point cloud between the obstacle and ground and slope surfaces in different urban scenarios. The segmentation accuracy can reach 98% on average, and the average time consumed is 2 ms.