Constrained Multiple Planar Reconstruction for Automatic Camera Calibration of Intelligent Vehicles

Area of Science:

• Computer Vision
• Robotics
• Automotive Engineering

Background:

• Extrinsic camera calibration is crucial for intelligent vehicles, requiring frequent updates due to mechanical variations.
• High-precision camera extrinsic parameters are vital for functions like distance estimation and lane departure warning.
• Conventional calibration methods (Perspective-n-Point) are labor-intensive due to manual 3D point measurement.

Purpose of the Study:

• To propose an automatic and convenient camera calibration method for intelligent vehicles.
• To develop a novel 3D reconstruction technique for recovering 3D points on ground-perpendicular planes.
• To significantly reduce errors in camera extrinsic parameters for enhanced vehicle safety systems.

Main Methods:

• A novel 3D reconstruction method is proposed to recover 3D points on planes perpendicular to the ground.
• The method jointly optimizes reprojection errors from multiple planar surfaces.
• This approach automates the calibration process, reducing manual effort.

Main Results:

• The proposed method effectively reconstructs 3D points on vertical planes.
• Joint optimization of reprojection errors leads to significant reduction in camera extrinsic parameter errors.
• Experiments in simulation and real-world environments validate the method's effectiveness.

Conclusions:

• The developed automatic camera calibration method offers a more convenient and accurate solution for intelligent vehicles.
• The novel 3D reconstruction technique addresses limitations of conventional calibration methods.
• This advancement contributes to more reliable and precise operation of intelligent vehicle safety features.