MFCalib: Single-shot and Automatic Extrinsic Calibration for LiDAR and Camera in Targetless Environments Based on Multi-Feature Edge

TL;DR

MFCalib tackles LiDAR-camera extrinsic calibration in targetless environments with a single data capture. It fuses three edge feature types—depth-continuous, depth-discontinuous, and intensity-discontinuous edges—and incorporates a LiDAR beam model to mitigate edge inflation, formulating a Gauss-Newton optimization in the tangent space of $SE(3)$. The method initializes from CAD priors and supports both static and rotating LiDARs, showing robustness across indoor and outdoor scenes. Experimental results on the SZU campus and Yuan's public dataset demonstrate superior accuracy and robustness compared to state-of-the-art targetless methods, with single-scene performance often matching or exceeding multi-scene calibrations, and the authors provide open-source code.

Abstract

This paper presents MFCalib, an innovative extrinsic calibration technique for LiDAR and RGB camera that operates automatically in targetless environments with a single data capture. At the heart of this method is using a rich set of edge information, significantly enhancing calibration accuracy and robustness. Specifically, we extract both depth-continuous and depth-discontinuous edges, along with intensity-discontinuous edges on planes. This comprehensive edge extraction strategy ensures our ability to achieve accurate calibration with just one round of data collection, even in complex and varied settings. Addressing the uncertainty of depth-discontinuous edges, we delve into the physical measurement principles of LiDAR and develop a beam model, effectively mitigating the issue of edge inflation caused by the LiDAR beam. Extensive experiment results demonstrate that MFCalib outperforms the state-of-the-art targetless calibration methods across various scenes, achieving and often surpassing the precision of multi-scene calibrations in a single-shot collection. To support community development, we make our code available open-source on GitHub.

Figures (10)

• Figure 1: The RGB-colored point cloud using the proposed method.
• Figure 2: System overview of MFCailb.
• Figure 3: Different kinds of edges in a real-world scene: Red lines mark depth-discontinuous edges, blue lines mark depth-continuous edges, and green lines mark intensity-discontinuous edges on plane.
• Figure 4: Inflation points and bleeding points caused by the divergence angle of LiDAR laser beam:(a) the actual inflation and bleeding points, where (1) is the point cloud colored by intensity values, and (2) is the intensity image; (b) the schematic of the LiDAR beam.
• Figure 5: Projection of an edge point from the LiDAR, denoted as $^C\mathbf{P}_i$ in the camera frame, onto the image plane $\mathbf{p}_i$. This includes calculating the residual $\mathbf{z}_i$, associated with the projection. LiDAR edges are marked in blue, and camera edges in red.