An Automated Extrinsic Calibration Method of a multi-layer LiDAR and a Camera

This article introduces an innovative technique for automating the extrinsic calibration of LiDAR and camera setups, employing a specially crafted circular calibration target. The process consists of three key phases: initially, the central characteristic and the perpendicular plane of the calibration target are identified within both the point cloud data and the captured imagery. Subsequently, constraint equations are established from these matched features to compute a preliminary estimation of the transformation matrix linking the LiDAR and the camera. In the final phase, a symmetric and nonnegative latent factor model enhanced by triple decomposition is applied to refine the extrinsic calibration to a high degree of precision. Empirical evaluations indicate a 47.59% enhancement in the alignment accuracy of 3D points from the LiDAR coordinate system with those of the image pixel coordinates relative to conventional methods. Notably, this performance improvement persists even under noise conditions reaching up to 0.1 dB, with the maximum reprojection error constrained below 9.35 pixels, thereby underscoring the method's resilience to noise interference.