Answers are grounded in this paper and cite their sources.
Accurate extrinsic calibration is essential for combining 3D LiDAR measurements with camera images, especially when sparse laser sampling complicates feature extraction. This study proposes a low-cost target-based method for 3D LiDAR and monocular or stereo camera calibration. A passive planar target with six circular holes provides corresponding LiDAR and image features. Both variants share the acquisition arrangement and LiDAR feature extraction procedure. Initial transformations are estimated from 3D-2D correspondences for the monocular variant and triangulated 3D-3D correspondences for the stereo variant, followed by refinement minimizing image reprojection errors. Experiments with a Velodyne VLP-16 and ZED2 stereo camera included ten independent calibration datasets and analyses of target distribution. Mean rotational deviations across repeated calibrations were 0.059° and 0.052° for the monocular and stereo variants, respectively. Their transformations differed by 0.030° in rotation and 2.0 mm in translation on average. Single-position calibration produced large errors at excluded target positions, whereas configurations spanning multiple distances yielded lower validation errors. Six positions covering the nearest and farthest distances achieved a mean validation RMSE of 3.58 pixels at the excluded intermediate distance. These findings demonstrate close correspondence between calibration variants and highlight the importance of spatial target distribution.
No takes yet. Share an insight, caveat, or question.
Brindza et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: