Multi-Sensor Extrinsic Calibration Using an Extended Set of Pairwise Geometric Transformations.

Area of Science:

• Robotics and Sensor Fusion
• Computer Vision
• Geometric Calibration

Background:

• Multi-sensor systems for exteroceptive perception are increasingly common in robotics and monitored spaces.
• Accurate sensor registration (knowing relative geometric transformations) is crucial for fusing data and creating robust scene representations.
• Traditional sensor calibration often requires significant human intervention, posing a challenge.

Purpose of the Study:

• To propose a nearly automatic method for calibrating multiple sensors by optimizing geometric transformations.
• To enhance the precision of sensor calibration by combining pairwise transformations and mitigating outliers.
• To provide a robust calibration solution applicable to various sensor types.

Main Methods:

• An experimental method generates individual pairwise transformations between sensors using 3D point clouds of a moving ball.
• A novel approach processes and combines these pairwise transformations, applying outlier rejection criteria.
• The method leverages the combination of transformations to achieve a more precise overall calibration.

Main Results:

• The proposed method significantly improves calibration accuracy compared to previous approaches that do not handle outliers.
• The multi-pairwise technique demonstrates enhanced geometric transformation precision by combining individual sensor data.
• The method is versatile, applicable to LIDAR, mono cameras, stereo cameras, and Time-of-Flight (TOF) cameras.

Conclusions:

• The developed method offers a nearly automatic and more accurate solution for multi-sensor calibration.
• Combining pairwise transformations effectively reduces uncertainty and improves the robustness of the perceived scene representation.
• This approach advances the field of sensor fusion by enabling more reliable data integration in complex systems.