Vehicle LiDAR-Camera Alignment Using Two-Stage Recalibration

The calibration process is divided into two distinct stages: pre-alignment and deep-alignment. The pre-alignment stage uses sparse data to establish initial calibration parameters, while the deep-alignment stage uses dense data to refine these parameters. This segmentation allows the system to achieve high alignment accuracy without requiring a single complex alignment procedure, as each stage builds upon the previous one incrementally.

The pre-alignment procedure serves as a preliminary action that prepares the calibration parameters before the final deep-alignment process. By first establishing rough alignment using sparse data collected during normal driving, the system creates a foundation that enables the subsequent deep-alignment to converge faster and achieve higher precision with dense data.

The data collection and processing is segmented into two phases: sparse data collection for pre-alignment and dense data collection for deep-alignment. This allows the system to use minimal data for initial calibration, reducing time loss, while reserving dense data collection for the final refinement stage where maximum accuracy is required.

Sparse data collection and pre-alignment serve as preliminary actions that reduce the burden on the deep-alignment stage. By establishing initial calibration parameters using sparse data, the system reduces the complexity and time required for processing dense data in the subsequent deep-alignment phase, as the optimization starts from a better initial state.

The system performs self-calibration using data collected from its own sensors during normal vehicle operation. The calibration process utilizes LiDAR and camera data that are already being collected for environmental perception, eliminating the need for external specialized calibration equipment. The system serves its own calibration needs by processing its operational data through the two-stage alignment procedure.

The calibration system leverages the existing LiDAR and camera sensors that are already installed for primary functions like obstacle detection and environmental mapping. These sensors serve dual purposes: their primary function for safety and navigation, and their data is simultaneously used for calibration. This multi-functionality eliminates the need for dedicated calibration equipment, reducing system complexity while maintaining alignment precision.