Relative pose estimation for multiple cameras using Lie algebra optimization.

Area of Science:

• Computer Vision and Robotics
• 3D Geometry and Calibration

Background:

• Accurate camera calibration is crucial for 3D reconstruction and augmented reality.
• Existing methods often require common features or targets across multiple cameras, limiting flexibility.
• Motion-based calibration offers an alternative but faces challenges in robustness and accuracy.

Purpose of the Study:

• To develop a robust framework for multi-camera relative pose estimation and calibration.
• To enable calibration of multiple cameras without requiring common spatial targets or features.
• To achieve high accuracy in camera pose estimation using limited camera-target pose information.

Main Methods:

• Formulating relative pose estimation from a set of camera-target relative poses.
• Developing a least-squares loss function utilizing 3D relative Riemannian metrics for cameras and targets.
• Employing Lie algebra-based solvers for efficient orientation and pose estimation.

Main Results:

• Successfully calibrated multiple cameras without relying on common targets or features.
• Demonstrated high accuracy in relative pose estimation through simulations and real-world experiments.
• Showcased the effectiveness of the proposed algorithms with limited and precise camera-target pose data.

Conclusions:

• The proposed framework provides an accurate and flexible solution for multi-camera calibration.
• Lie algebra-based methods offer efficient and precise solutions for relative pose estimation.
• The approach is effective even with sparse and accurate camera-target pose measurements.