How do you achieve and maintain accurate lidar-camera extrinsic calibration?

Lidar-camera calibration: 1) Initial calibration - using calibration targets (checkerboard with retroreflective markers, spherical targets), compute transformation minimizing reprojection error of corresponding points; 2) Target-less calibration - exploiting natural features (edges, planes) visible in both modalities, mutual information or edge-based correspondence; 3) Online refinement - continuous monitoring and adjustment for thermal drift, vibration-induced changes using feature correspondence; 4) Validation - projection verification on static scenes, cross-checking object positions. Accuracy requirements are sub-degree rotation, centimeter translation. Challenges include different sensor characteristics (lidar sparse, camera dense), limited overlap at close range, and different update rates. Robust calibration is critical for accurate fusion, particularly for 3D object localization.