Targetless 6DoF Calibration of LiDAR and 2D Scanning Radar Based on Cylindrical Occupancy

TL;DR

LiRaCo addresses the challenge of targetless 6DoF extrinsic calibration between LiDAR and 2D scanning Radar by constructing 3D occupancy grids from Radar data in a cylindrical coordinate frame and aligning them with LiDAR points through a spatial occupancy consistency cost. The method expands Radar scans to 3D grids, employs a vertical restrainer and an intensity factor to handle height sparsity and reflectivity, and optimizes the extrinsics with Trust Region Reflective. Experimental validation on ORR and Boreas datasets shows that LiRaCo achieves accurate, robust calibration with all six parameters, outperforming traditional baselines and demonstrating stability against initialization and radar intensity variations. This approach enables robust, marker-free sensor fusion for all-weather autonomous perception in real-world scenarios.

Abstract

Owing to the capability for reliable and all-weather long-range sensing, the fusion of LiDAR and Radar has been widely applied to autonomous vehicles for robust perception. In practical operation, well manually calibrated extrinsic parameters, which are crucial for the fusion of multi-modal sensors, may drift due to the vibration. To address this issue, we present a novel targetless calibration approach, termed LiRaCo, for the extrinsic 6DoF calibration of LiDAR and Radar sensors. Although both types of sensors can obtain geometric information, bridging the geometric correspondences between multi-modal data without any clues of explicit artificial markers is nontrivial, mainly due to the low vertical resolution of scanning Radar. To achieve the targetless calibration, LiRaCo leverages a spatial occupancy consistency between LiDAR point clouds and Radar scans in a common cylindrical representation, considering the increasing data sparsity with distance for both sensors. Specifically, LiRaCo expands the valid Radar scanned pixels into 3D occupancy grids to constrain LiDAR point clouds based on spatial consistency. Consequently, a cost function involving extrinsic calibration parameters is formulated based on the spatial overlap of 3D grids and LiDAR points. Extrinsic parameters are finally estimated by optimizing the cost function. Comprehensive quantitative and qualitative experiments on two real outdoor datasets with different LiDAR sensors demonstrate the feasibility and accuracy of the proposed method. The source code will be publicly available.

Figures (9)

• Figure 1: The challenge for LiDAR-Radar calibration and proposed targetless approach. Left: the LiDAR and Radar data of a scene with the RGB image for reference, we can find the difference in physical significance between two sensors in terms of reflective signal and intensity; right: we expand Radar points to 3D areas in cylindrical coordinate system to construct spatial constraints.
• Figure 2: The overview of our proposed LiRaCo for targetless LiDAR-Radar extrinsic calibration. Both LiDAR and Radar are represented in cylindrical coordinates. Radar pixels are expanded to 3D occupancy grids to constrain LiDAR points that are transformed with initial or estimated extrinsic parameters.
• Figure 3: The illustration of the generated Radar 3D occupancy grid in the cylindrical coordinate system and the reference Cartesian view, considering the form of the Radar wave. The center of the grid is set at $\textbf{s}^{3D}_i=(r_i, \theta_i, 0, i_i)$ and the size of the grid is defined as $(\Delta r,~\Delta \beta_H,~h_i^f,~h_i^r)$ which depend on the range resolution, vertical and horizontal angular resolution and the distance to the coordinate origin.
• Figure 4: The illustration of the height restrainer. LiDAR points are expected to be transformed towards the center of the Radar occupancy grid.
• Figure 5: Initial parameters and the estimated results for each extrinsic in (a) ORR and (b) Boreas Dataset. The X-axis represents the index of the trial and the Y-axis is the value of the parameters in $[\theta]$ and $[cm]$ respectively. Light blue square indicates the initial value and the optimized result is in dark blue point. The red line shows the GT.