NFL-BA: Near-Field Light Bundle Adjustment for SLAM in Dynamic Lighting

Area of Science:

• Robotics and Computer Vision
• Artificial Intelligence
• Medical Imaging

Background:

• Simultaneous Localization and Mapping (SLAM) systems often fail in dynamic, near-field lighting conditions due to view-dependent shading.
• Real-world applications like endoscopy and subterranean robotics require robust SLAM in the absence of external lighting.

Purpose of the Study:

• To develop a novel loss function, Near-Field Lighting Bundle Adjustment Loss (NFL-BA), for enhancing SLAM performance under dynamic, near-field lighting.
• To integrate NFL-BA into neural rendering-based SLAM systems for improved camera tracking and mapping.

Main Methods:

• Explicitly modeling near-field lighting within the Bundle Adjustment loss function.
• Integrating NFL-BA into existing SLAM frameworks, such as MonoGS and EndoGS.
• Evaluating performance on colonoscopy datasets (C3VD) and indoor scenes with on-camera flash.

Main Results:

• Significant improvements in camera tracking accuracy (37% for MonoGS, 14% for EndoGS) compared to traditional Photometric Bundle Adjustment loss.
• Achieved state-of-the-art camera tracking and mapping performance on the C3VD colonoscopy dataset.
• Demonstrated substantial SLAM performance gains in indoor scenes with on-camera flash.

Conclusions:

• NFL-BA effectively addresses the challenges posed by dynamic near-field lighting in SLAM.
• The proposed method offers significant benefits for autonomous navigation and 3D visualization in endoscopic procedures.
• NFL-BA provides a robust solution for improving SLAM accuracy across various applications with challenging lighting conditions.