Multi-Cue-Based Circle Detection and Its Application to Robust Extrinsic Calibration of RGB-D Cameras

Area of Science:

• Computer Vision
• Robotics
• 3D Sensing

Background:

• Multiple RGB-Depth (RGB-D) cameras are crucial for 3D modeling and human-computer interaction.
• Sparse camera distribution without shared features presents challenges for extrinsic calibration.
• Spherical objects are valuable for calibrating widely-separated cameras due to their visibility from multiple viewpoints.

Purpose of the Study:

• To develop a robust method for detecting circular regions in single color images, specifically for identifying spherical objects.
• To apply this detection method to the extrinsic calibration of multiple RGB-D cameras.
• To improve calibration accuracy and robustness, especially in the presence of outliers and challenging environmental conditions.

Main Methods:

• A multi-cue-based algorithm for detecting circular regions in color images.
• Utilizing detected circular regions (spherical objects) for extrinsic camera calibration.
• Implementing robust cost functions to mitigate errors from misdetected sphere centers.

Main Results:

• The proposed circle detection method accurately identifies spherical objects in cluttered backgrounds and varying illumination.
• The extrinsic calibration method using robust cost functions demonstrates superior performance compared to least-squares methods.
• Accurate calibration results are achieved even with outliers, indicating enhanced robustness.

Conclusions:

• The developed multi-cue circle detection is effective for identifying spherical objects in challenging conditions.
• The proposed robust cost function approach significantly improves the extrinsic calibration of multiple RGB-D cameras.
• This method offers a more accurate and reliable solution for calibrating distributed camera systems.