Automatic Extrinsic Calibration of 3D LIDAR and Multi-Cameras Based on Graph Optimization

Area of Science:

• Robotics and Computer Vision
• Sensor Fusion and Calibration

Background:

• Extrinsic calibration is crucial for multi-sensor fusion in applications like autonomous driving and 3D reconstruction.
• Existing calibration methods face challenges in accuracy and robustness.

Purpose of the Study:

• To propose an automatic 3D LIDAR-to-camera calibration framework.
• To simultaneously calibrate LIDAR and multiple cameras.
• To improve calibration accuracy and robustness.

Main Methods:

• A novel framework based on graph optimization for automatic LIDAR-to-camera calibration.
• Automatic identification of calibration patterns and creation of virtual feature point clouds.
• Simultaneous calibration of LIDAR with monocular and binocular cameras.

Main Results:

• Achieved an average error of 0.161 mm on the camera normalization plane.
• Demonstrated superior accuracy compared to state-of-the-art methods.
• The graph optimization simultaneously refines point clouds, correcting data collection errors and enhancing robustness.

Conclusions:

• The proposed graph optimization-based framework provides accurate and robust automatic calibration for LIDAR-camera systems.
• This method is effective for multi-sensor fusion applications, including autonomous driving and 3D mapping.
• The framework's ability to optimize point clouds makes it resilient to noisy data.