A Full-Lifecycle Calibration Method for Camera and LiDAR in Autonomous Driving

Engineering notes

This full-lifecycle calibration method significantly improves the accuracy, reliability, and long-term stability of autonomous driving systems, addressing critical challenges in sensor calibration across diverse environments. The code will be released at https://github.com/weiming2/off-onlineCalib.git

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

In recent years, advancements in light laser detection and ranging (LiDAR) and camera technologies have brought increasing attention to autonomous driving. Sensor fusion provides complementary information that overcomes the limitations of individual sensors and improves the safety of autonomous vehicles. The key to achieving this fusion lies in sensor calibration. This article presents a full-lifecycle calibration method that integrates offline and online processes to achieve high-precision and efficient calibration for camera-LiDAR systems in autonomous driving. The proposed approach unifies the calibration stages, addressing both the initial factory calibration and the dynamic adjustments required during vehicle operation. Offline calibration is performed using specialized calibration boards and standardized workflows to establish accurate initial parameters, ensuring a robust foundation for subsequent operations. Online calibration leverages a learning-based, end-to-end deep declarative network to dynamically adjust calibration parameters in real time, compensating for sensor displacement caused by vibration, loosening, or collisions. Extensive experiments are conducted to validate the proposed method. The accuracy and robustness of the offline calibration are validated through quantitative and qualitative results in simulated and real-world tests. Competitive precision in online calibration is demonstrated on public datasets, with average translation errors of 0.667 and 1.937 cm, and average rotation errors of 0.149° and 0.138° achieved on the KITTI and NuScenes datasets, respectively. Additionally, the online calibration method’s real-time capability is confirmed in practical experiments, with a frame rate of approximately 11 frames/s. This full-lifecycle calibration method significantly improves the accuracy, reliability, and long-term stability of autonomous driving systems, addressing critical challenges in sensor calibration across diverse environments. The code will be released at https://github.com/weiming2/off-onlineCalib.git

Links and sources

• Official / arXiv page

Comments