Ground Segmentation of 3D LiDAR Point Cloud with Adaptive Threshold

LiDAR has been widely used in autonomous vehicles for its advantages of wide horizon and accurate detection. However, there are still several challenges in processing LiDAR point cloud, where the ground segmentation is one of the critical issues. In complex driving scenarios such as slope terrain and road bump, the current ground segmentation methods have the problem of under-segmentation or over-segmentation and its real-time performance degradation becomes unacceptable. Motivated by this, a 3D LiDAR point cloud ground segmentation algorithm based on adaptive ground height and adaptive slope threshold is proposed to segment the ground points in complex terrain. First, the point cloud is filtered by voxelization method, and the cylindrical sector is constructed with the help of polar coordinate. Then the point cloud is divided into different sectors, and the slope threshold and ground height are updated in each sector. Finally, an adaptive ground height and an adaptive slope threshold are used to distinguish ground points and non-ground ones. Qualitative and quantitative experiments on both SemanticKITTI dataset and on-board sensor data are conducted to evaluate the accuracy and efficiency of our algorithm.