CAR-LOAM: Color-Assisted Robust LiDAR Odometry and Mapping

TL;DR

This paper addresses robust LiDAR odometry and mapping under challenging conditions by exploiting color information from cameras to colorize LiDAR data. The method CAR-LOAM integrates color-based color weighting using $\Delta E_{00}^*$ in $L^*C^*H^*$ space and a Welsch-based robust loss into a feature-based LOAM framework for scan-to-map matching. Key contributions include dense colored 3D map reconstruction, color-assisted robust LOAM without extra visual pose module, and comprehensive experiments demonstrating improved robustness and accuracy in forest, campus, and building reconstruction. The work advances LOAM robustness for multi-modal sensing in robotics with practical implications for real-world navigation and mapping.

Abstract

In this letter, we propose a color-assisted robust framework for accurate LiDAR odometry and mapping (LOAM). Simultaneously receiving data from both the LiDAR and the camera, the framework utilizes the color information from the camera images to colorize the LiDAR point clouds and then performs iterative pose optimization. For each LiDAR scan, the edge and planar features are extracted and colored using the corresponding image and then matched to a global map. Specifically, we adopt a perceptually uniform color difference weighting strategy to exclude color correspondence outliers and a robust error metric based on the Welsch's function to mitigate the impact of positional correspondence outliers during the pose optimization process. As a result, the system achieves accurate localization and reconstructs dense, accurate, colored and three-dimensional (3D) maps of the environment. Thorough experiments with challenging scenarios, including complex forests and a campus, show that our method provides higher robustness and accuracy compared with current state-of-the-art methods.

Figures (13)

• Figure 1: (a) A dense, accurate, colored and 3D point cloud map reconstructed by CAR-LOAM. (b) A close-up of (a), highlighting autumn trees and a car beneath them. (c) Another close-up of (a), showing one side of the building, including its doors and windows. (d) A corresponding satellite image of the campus environment, with the red path indicating the traveling trajectory of our mobile platform.
• Figure 2: The overview of our workflows. Each new LiDAR frame, colored by the corresponding RGB color image from the camera, is matched with the global map to provide the odometry output. The matching result is in turn used to register the frame to the global map. During the matching process, each point on the new colored edge/planar feature is used to calculate the distance and perceptually uniform color difference to its corresponding edge/planar feature retrieved from the global map. The color of the edge/planar feature is represented by the color of the nearest neighbor point of the current query point. The distances serve as inputs to the Welsch's function to get robust normalized residuals. Similarly, perceptually uniform color differences are processed using the Gaussian robust function to generate normalized color weights. The residuals are then multiplied by their corresponding weights and utilized in iterative pose optimization.
• Figure 3: Steps of the coloring of LiDAR clouds and features.
• Figure 4: The Welsch's function $\psi_\nu$ with $\nu=2$ and the Gaussian robust function $\rho_\sigma$ with $\sigma=5$.
• Figure 5: Left: mapping results in Forest01 by CAR-LOAM. The yellow path is the corresponding trajectory. Right: Trajectories in Forest01.