A Hierarchical Scheme of Road Unevenness Perception With LiDAR for Autonomous Driving Comfort

Road unevenness is a significant factor affecting the comprehensive driving performance of vehicles. Monitoring road roughness conditions in advance significantly contributes to ride safety and comfort. However, there are hardly comprehensive implementations for comfort promotion purpose based on LiDAR point clouds. This paper proposes a hierarchical scheme that separately describes road regular and irregular unevenness. Firstly, real-vehicle experiments for collecting point clouds of various uneven roads are conducted with the designed hardware platform. A dense point cloud dataset with road bump or pothole annotations is released. Then, an improved PointNet++ model adapted to irregular unevenness recognition is developed. Comparison experiments with the recent segmentation methods verify the superiority of our model in this task. The mIoU metric on the test set reaches 91.6%, outperforming 3.0% than the original model. Road elevation profiles on tire trajectories are finally constructed. The statistical parameters of regular unevenness and geometric features of the irregular sections are extracted, respectively. The proposed perception scheme for road roughness provides detailed and accurate road condition information. It has great potential in practical applications for road unevenness monitoring and vehicle comfort improvement.