Online LiDAR-to-Vehicle Alignment Using Lane Markings and Traffic Signs

Yao Hu, Xinyu Du, Shengbing Jiang; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2023, pp. 3348-3357

Abstract


Highly automated vehicles with multiple environmental sensors require all the sensors aligned online to the same coordinate to ensure driving performance and improve customer convenience, especially when misalignment occurs during driving due to degradation, ageing, vibration, or accidents. The alignment between the LiDAR and the ego vehicle is one of several types of alignments. In this paper, an online alignment approach using road elements, e.g., lane markings and traffic signs, in aggregated LiDAR point cloud is developed. The optimization process to minimize the variance of aggregated point cloud for each road element is employed to automatically calculate the alignment parameters. To improve the algorithm robustness and accuracy, several excitation conditions occurred in daily driving are identified by algorithm sensitivity analysis with small input perturbations. The road elements are detected using unique designed heuristic algorithms from the distorted point cloud due to the inaccurate alignment parameters during optimization. The whole solution is validated by the data collected from several test vehicles, and the validation results demonstrate the effectiveness and robustness of the proposed solution.

Related Material


[pdf]
[bibtex]
@InProceedings{Hu_2023_CVPR, author = {Hu, Yao and Du, Xinyu and Jiang, Shengbing}, title = {Online LiDAR-to-Vehicle Alignment Using Lane Markings and Traffic Signs}, booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops}, month = {June}, year = {2023}, pages = {3348-3357} }