ICET Online Accuracy Characterization for Geometry-Based Laser Scan Matching

TL;DR

The Iterative Closest Ellipsoidal Transform (ICET) is introduced, a novel 3D LIDAR scan-matching algorithm that re-envisions NDT in order to provide robust accuracy prediction from first principles and is well suited for safety-critical transportation applications.

Abstract

Distribution-to-Distribution (D2D) point cloud registration algorithms are fast, interpretable, and perform well in unstructured environments. Unfortunately, existing strategies for predicting solution error for these methods are overly optimistic, particularly in regions containing large or extended physical objects. In this paper we introduce the Iterative Closest Ellipsoidal Transform (ICET), a novel 3D LIDAR scan-matching algorithm that re-envisions NDT in order to provide robust accuracy prediction from first principles. Like NDT, ICET subdivides a LIDAR scan into voxels in order to analyze complex scenes by considering many smaller local point distributions, however, ICET assesses the voxel distribution to distinguish random noise from deterministic structure. ICET then uses a weighted least-squares formulation to incorporate this noise/structure distinction into computing a localization solution and predicting the solution-error covariance. In order to demonstrate the reasonableness of our accuracy predictions, we verify 3D ICET in three LIDAR tests involving real-world automotive data, high-fidelity simulated trajectories, and simulated corner-case scenes. For each test, ICET consistently performs scan matching with sub-centimeter accuracy. This level of accuracy, combined with the fact that the algorithm is fully interpretable, make it well suited for safety-critical transportation applications. Code is available at https://github.com/mcdermatt/ICET

Figures (10)

• Figure 1: Voxel point distributions featuring (A) sensor noise, (B) surface roughness, and (C) extended surfaces
• Figure 2: Illustration of a wall (seen from above) passing through two rounded voxels. LIDAR measurements of the surface are shown as dots. The mean and covariance matrix for the scan points in each voxel are indicated by the shaded ellipsoids.
• Figure 3: Sigma-point refinement for identification of extended surfaces within a single voxel. A covariance ellipse is shown in relation to a wireframe voxel. The sigma points along the ellipsoid's three principal axes (labeled I, II, and III), are identified with small dots. Only axis III is eliminated by our sigma-point exclusion text, since both of its sigma points lie outside the voxel boundaries.
• Figure 4: Spherical voxels with adaptive radial binning. Point distributions within each cell are shown as ellipsoids.
• Figure 5: Synced forward facing images from LeddarTech PixSet (pixset), drive 20200721_144638_part36_1956_2229 through Old Montreal, Quebec.