Visualization of MULi-Ev results across different driving environments
Abstract
Online extrinsic calibration between event cameras and LiDAR has not been addressed in prior work, despite both sensors appearing on research AV platforms.
We present MULi-Ev, an online deep learning method for LiDAR–event calibration. It adjusts extrinsics during driving without manual targets or offline recalibration.
On the DSEC dataset, MULi-Ev improves calibration accuracy over baseline methods and supports LiDAR–event fusion on moving platforms.
Key Contributions
- Online LiDAR–event extrinsics: estimates the 6-DoF transform while driving, without chessboards or offline batch recalibration
- Event-aware architecture: designed for the sparse, asynchronous measurements of event cameras rather than frame-based RGB
- DSEC evaluation: improves calibration accuracy over baselines and supports downstream LiDAR–event fusion on moving platforms
- AV-ready setting: targets research autonomous-vehicle stacks where LiDAR and event cameras co-exist
Citation
@InProceedings{Cocheteux_2024_CVPR,
author = {Cocheteux, Mathieu and Moreau, Julien and Davoine, Franck},
title = {MULi-Ev: Maintaining Unperturbed LiDAR-Event Calibration},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2024},
pages = {4579-4586}
}