Automatic LiDAR-Camera Extrinsic Calibration using Pseudo Image and Multiple Targets

Recently, with the rapid progress of Solid-State LiDAR (SSL), the issues of establishing a robust transformation between SSL and other sensors have attracted considerable attention. However the inconsistency of the ranging error distribution and the trailing phenomenon of SSL pose great challenges to the calibration. To address this issue, the article proposes an automatic calibration method based on pseudo-visual images obtained from SSL and multiple targets with distinguished patterns. For feature refinement, by ultilizing adaptive pseudo-visual image restoration algorithms, the precise detection of checkerboard corners in point clouds brings reliable initial values for coarse registration. The biplanar fitting and hollow fillment algorithms are invented to efficiently extract the ideal planes from noisy point clouds for precise registration. For registration, the 2D-3D corner mapping table is established to estimate the initial rigid-body transformation. After coarse registration, the loss function based on multiplanar distances and normal vectors is introduced to obtain the final extrinsic parameters in precise registration. Our proposed method has been tested in real-world and synthetic environments with extensive qualitative and quantitative analysis respectively, to verify its excellent accuracy and robustness compared with other SOTAs.