Visual-Lidar Map Alignment for Infrastructure Inspections

TL;DR

The paper tackles long-term infrastructure health monitoring in GPS-denied environments, where aligning 3D maps from repeated inspections is essential for accurate degradation tracking. It introduces a SLAM-agnostic Visual-Lidar Map Alignment (VLMA) pipeline that fuses visual place recognition and lidar place recognition to robustly identify map correspondences across sessions, followed by map-to-map scan registration with Generalized-ICP and a non-rigid trajectory alignment built on a $\mathcal{B}$-Spline to account for drift. Key contributions include the modular VLMA framework, a non-rigid alignment formulation that maximizes local map overlap, and extensive experiments on indoor and outdoor datasets showing VLMA’s superior robustness to single-modality methods. The approach enables independent SLAM developments to produce aligned maps in a common reference frame, facilitating scalable, automated infrastructure inspections and long-term asset health assessment, with publicly available code for broader adoption and replication.

Abstract

Routine and repetitive infrastructure inspections present safety, efficiency, and consistency challenges as they are performed manually, often in challenging or hazardous environments. They can also introduce subjectivity and errors into the process, resulting in undesirable outcomes. Simultaneous localization and mapping (SLAM) presents an opportunity to generate high-quality 3D maps that can be used to extract accurate and objective inspection data. Yet, many SLAM algorithms are limited in their ability to align 3D maps from repeated inspections in GPS-denied settings automatically. This limitation hinders practical long-term asset health assessments by requiring tedious manual alignment for data association across scans from previous inspections. This paper introduces a versatile map alignment algorithm leveraging both visual and lidar data for improved place recognition robustness and presents an infrastructure-focused dataset tailored for consecutive inspections. By detaching map alignment from SLAM, our approach enhances infrastructure inspection pipelines, supports monitoring asset degradation over time, and invigorates SLAM research by permitting exploration beyond existing multi-session SLAM algorithms.

Figures (8)

• Figure 1: Map alignment pipeline starting from input (a set of images, point clouds and trajectory), to visual place recognition (VPR) and lidar place recognition (LPR) modules which propose matches between maps, to relative pose estimate and finally map alignment to produce the output (a merged point cloud map).
• Figure 2: Visual place recognition flow chart, starting from the sets of images in each map (left), building and querying the database, and finally verifying matches.
• Figure 3: Starting map pair with correspondences (left), rigid map alignment (top right) and non-rigid map alignment (bottom right).
• Figure 4: Data collection device outfitted with Velodyne lidar, 3 Flir cameras and Xsens IMU.
• Figure 5: Image from Conestogo Bridge dataset (left) along with the 3D map (right).