Adaptive Point-Line Fusion: A Targetless LiDAR-Camera Calibration Method with Scheme Selection for Autonomous Driving

Area of Science:

• Robotics and Computer Vision
• Sensor Fusion for Autonomous Systems

Background:

• Accurate LiDAR-camera calibration is vital for autonomous driving perception.
• Traditional methods require overlapping fields of view and struggle with feature alignment due to noise.

Purpose of the Study:

• To develop a targetless extrinsic calibration method for monocular cameras and LiDAR with non-overlapping fields of view.
• To enhance the robustness and accuracy of sensor calibration in diverse urban driving conditions.

Main Methods:

• Pose transformation for cross-modal data association.
• Visual SLAM optimization framework without overlapping views.
• Line features as constraints with extended photometric error optimization.
• Adaptive selection strategy for calibration methods.

Main Results:

• Successful targetless extrinsic calibration for non-overlapping LiDAR-camera sensors.
• Improved calibration robustness in varying urban lighting and textures.
• Demonstrated avoidance of failures and bias associated with single calibration approaches.

Conclusions:

• The proposed method offers a robust and adaptable solution for LiDAR-camera extrinsic calibration.
• This approach significantly advances perception capabilities for autonomous driving systems.
• Adaptive method selection ensures reliable performance across challenging real-world scenarios.