Laser-to-Vehicle Extrinsic Calibration in Low-Observability Scenarios for Subsea Mapping

TL;DR

This work tackles laser-to-vehicle extrinsic calibration in challenging subsea environments where navigation observability can be limited. It develops three optimization-based algorithms operating on $SE(3)$ with a shared Tikhonov regularization to address low-observability, leveraging natural 3D features instead of calibration targets. The methods are validated on two field datasets (Wiarton shipwreck and Endurance wreck), with Algorithm 2 consistently delivering the best map quality by jointly refining extrinsics and submap poses. The results demonstrate centimeter-level accuracy improvements and enable patch-test–free calibration for high-resolution subsea mapping, enhancing robustness for rotationally stable offshore vehicles.

Abstract

Laser line scanners are increasingly being used in the subsea industry for high-resolution mapping and infrastructure inspection. However, calibrating the 3D pose of the scanner relative to the vehicle is a perennial source of confusion and frustration for industrial surveyors. This work describes three novel algorithms for laser-to-vehicle extrinsic calibration using naturally occurring features. Each algorithm makes a different assumption on the quality of the vehicle trajectory estimate, enabling good calibration results in a wide range of situations. A regularization technique is used to address low-observability scenarios frequently encountered in practice with large, rotationally stable subsea vehicles. Experimental results are provided for two field datasets, including the recently discovered wreck of the Endurance.

Figures (8)

• Figure 1: Patch test scans of the Endurance shipwreck, with point disparity errors Roman2006 shown before and after joint extrinsic and trajectory optimization. Included with permission from the Falklands Maritime Heritage Trust.
• Figure 2: The Insight Pro underwater line scanner by Voyis Imaging Inc., showing the approximate location of datum point $s$ and sensor reference frame ${\mathcal{F}_{\ell}}$. The baseline between the line projector (left) and the camera (right) is approximately 1m.
• Figure 3: Defining a reprojection error between matched keypoints $\mbf{r}^{q_1w}_{a}$ and $\mbf{r}^{q_2w}_{a}$. Vehicle poses are shown in black. The design variable is the laser-to-INS extrinsics.
• Figure 4: Defining a reprojection error $\mbf{e}_j$ for Algorithm 2, which allows for global submap drift. Comparison to \ref{['fig:reprojection_errors_1']} shows the individual vehicle poses (black) have been replaced with the central submap poses and rigid offsets. The design variables are the laser-to-INS extrinsics and the central submap poses.
• Figure 5: A depiction of Algorithm 3, with exteroceptive errors placed on the DVL-INS poses and interoceptive errors linking all DVL-INS and measurement poses, both shown in black. The design variables include the laser-to-INS extrinsics as well as the intrinsic and extrinsic variables.