Targetless LiDAR-Camera Extrinsic Calibration via Class-Agnostic Boundary Mask Alignment and SPSA-Based Optimization

Area of Science:

• Robotics
• Computer Vision
• Sensor Fusion

Background:

• Targetless extrinsic calibration for LiDAR-camera systems is difficult due to poor cross-modal correspondences and initialization sensitivity.
• Existing methods often struggle with robustness and efficiency in real-world scenarios.

Purpose of the Study:

• To develop a targetless LiDAR-camera extrinsic calibration framework that overcomes current limitations.
• To improve the accuracy and robustness of sensor alignment for autonomous systems.

Main Methods:

• A novel framework utilizing class-agnostic boundary mask alignment in a shared image-plane representation.
• Construction of consistent LiDAR-camera mask pairs via image-plane projections.
• Robust pose initialization using bounded rotation-only global optimization.
• Refinement of extrinsic parameters via computationally efficient stochastic gradient approximation.

Main Results:

• Demonstrated superior accuracy-runtime trade-off on the KITTI benchmark compared to segmentation-based global optimization.
• Confirmed stable cross-modal alignment under vibration and inter-modal timing jitter in real-world driving tests.

Conclusions:

• The proposed class-agnostic boundary mask alignment offers a robust and efficient solution for targetless LiDAR-camera extrinsic calibration.
• This method enhances the reliability of sensor fusion for autonomous driving applications.