In recent years, 3D lane detection has become one of the most crucial and challenging tasks in the field of autonomous driving. While most research focuses on real-time or onboard scenario, there is still limited exploration of using machines to automatically generate high-precision 3D lane labels. Existing 3D lane labeling approaches struggle in efficiency and accuracy due to the single frame interactive annotation paradigm or information restrictions. In this paper, we propose a compact road-reconstruction based 3D lane auto labeling method for autonomous driving, termed R2-3DLane. It involves Lidar point cloud semantic segmentation, lane instance generation and post-processing. Additionally, we design a coarse-to-fine multi-stage Lidar point cloud clustering procedure that converts the lane semantic information into the lane instance annotations. The evaluation results on Waymo open dataset prove the effectiveness of the proposed method.
A Compact 3D Lane Auto Labeling Method for Autonomous Driving
2023-10-13
1793058 byte
Conference paper
Electronic Resource
English
LANE RECOMMENDATION METHOD AND LANE RECOMMENDATION DEVICE FOR AUTONOMOUS DRIVING
European Patent Office | 2021
|Auto-Sapiens Autonomous Driving Vehicle
TIBKAT | 2020
|