Multi-Modal Information Integration with Segmentation Clustering for Robust LiDAR-Camera Calibration

To effectively acquire high-precision data, integrating multi-modal information from different sensors becomes crucial. LiDAR and cameras are typically among the most common sources of multi-modal data. When calibrating the two, concepts and techniques from mathematics and statistics are widely employed. Addressing the limitation of conventional extrinsic calibration methods in scenes with a single and highly similar structure, we propose a robust calibration approach for the external parameters of integrated solid-state LiDAR-camera fusion systems. Initially, based on a statistical model, we utilize an innovative point cloud clustering and segmentation algorithm to combine global scene structures with local convexity information to extract more comprehensive overall structural information. Building upon this foundation, we introduce a calibration algorithm for LiDAR-camera systems based on point cloud and image segmentation clustering. This algorithm performs label-driven matching post cluster segmentation, initializes calibration for corresponding segments, and further optimizes external parameters using edge feature detection. Experimental evaluations indicate that in various scenarios, the rotation and translation parameters converge, reducing the average residual to 0.83. After segmentation clustering, the average time was reduced by 73 seconds.