We introduce a novel architecture, UniCal, for Camera-to-LiDAR (C2L) extrinsic calibration which leverages self-attention mechanisms through a Transformer-based backbone to infer the 6-degree of freedom (DoF) relative transformation between the sensors.
Unlike previous methods, UniCal performs an early fusion of the input camera and LiDAR data by aggregating camera image channels and LiDAR mappings into a multi-channel unified representation before extracting their features jointly with a single-branch architecture. This lightweight design makes UniCal well-suited for resource-constrained applications such as autonomous driving.
Through experiments, we show that UniCal achieves state-of-the-art results compared to existing methods. We also demonstrate that through transfer learning, weights learned on the calibration task can be applied to a calibration validation task without re-training the backbone.
- Single-branch architecture: novel transformer model performing early fusion of camera and LiDAR data via a unified multi-channel representation
- Lightweight design: efficient architecture suitable for resource-constrained autonomous driving deployments
- State-of-the-art performance: superior calibration accuracy vs. existing deep learning methods
- Transfer learning to validation: backbone weights transfer from calibration to validation without re-training
- 6-DoF transformation: full relative transformation between camera and LiDAR inferred end-to-end
This research led to an international patent filed through WIPO (PCT), reflecting its industrial relevance for autonomous driving and robotics systems at scale.
@article{cocheteux2023unical,
title = {UniCal: a Single-Branch Transformer-Based Model for
Camera-to-LiDAR Calibration and Validation},
author = {Cocheteux, Mathieu and Low, Aaron and Bruehlmeier, Marius},
journal = {arXiv preprint arXiv:2304.09715},
year = {2023}
}