MCSfM: Multi-Camera-Based Incremental Structure-From-Motion.

Area of Science:

• Robotics
• Computer Vision
• 3D Reconstruction

Background:

• Autonomous robots require robust environmental perception for navigation and interaction.
• Structure from Motion (SfM) is crucial for 3D scene reconstruction using multi-camera systems.
• Existing incremental SfM methods face challenges with efficiency and drift in large-scale tasks.

Purpose of the Study:

• To develop a novel incremental SfM framework tailored for multi-camera systems.
• To enable automatic calibration of internal camera poses within the framework.
• To enhance system robustness and accuracy for large-scale 3D reconstruction.

Main Methods:

• A reference camera is established, with non-reference camera poses derived from relative poses.
• A two-stage camera registration module computes internal relative poses via local motion averaging.
• Incremental registration of rigid units and multi-camera bundle adjustment refine camera poses and internal calibrations.

Main Results:

• The proposed framework demonstrates superior accuracy and robustness compared to state-of-the-art SfM and SLAM methods.
• Automatic calibration of internal relative poses improves system performance.
• The system effectively handles arbitrary multi-camera configurations with only image input.

Conclusions:

• The developed SfM framework offers a more robust and accurate solution for multi-camera 3D scene reconstruction.
• This approach addresses limitations of traditional SfM in large-scale applications for autonomous systems.
• The method provides a significant advancement for robotic perception and mapping.