EdgeDenseCalib: Targetless Camera-LiDAR Calibration via Enhanced Edge Feature Densification.

Area of Science:

• Robotics and Autonomous Systems
• Computer Vision
• Sensor Fusion

Background:

• Accurate camera-LiDAR calibration is crucial for autonomous systems.
• Traditional methods require manual intervention or calibration targets, limiting real-world application.
• Targetless calibration remains a significant challenge.

Purpose of the Study:

• To develop an automatic and targetless camera-LiDAR calibration method.
• To enhance the comparability of sparse LiDAR edge features with dense image features.
• To improve the reliability and scalability of calibration for autonomous systems.

Main Methods:

• Proposed EdgeDenseCalib, a novel approach using enhanced edge feature densification.
• Implemented a two-stage process to densify sparse LiDAR edge features.
• Utilized an optimization algorithm to refine alignment and minimize reprojection error.

Main Results:

• Achieved accurate calibration with mean rotation error of 0.105° and mean translation error of 0.903 cm on the KITTI dataset.
• Significantly improved rotation accuracy by 33.1% to 89.9% compared to state-of-the-art edge-based methods.
• Demonstrated reliable feature matching between cross-modal data sources.

Conclusions:

• EdgeDenseCalib provides a practical and automatic solution for camera-LiDAR calibration.
• The method enhances perception system robustness for autonomous applications.
• This work contributes to advancing targetless calibration techniques.