Neural Radiance Fields-Based 3D Reconstruction of Power Transmission Lines Using Progressive Motion Sequence Images

Area of Science:

• Computer Vision
• Robotics
• Electrical Engineering

Background:

• Reconstructing power transmission lines (PTLs) presents challenges due to fuzzy distant object effects and difficulties in feature matching thin structures.
• Existing methods struggle with the unbounded nature of PTL scenes and require robust feature extraction.

Purpose of the Study:

• To propose a novel image-based method for accurate 3D reconstruction of power transmission lines (PTLs).
• To address limitations in current reconstruction techniques, particularly for distant objects and thin structures.

Main Methods:

• Introduced PL-NeRF, an enhanced Neural Radiance Fields (NeRF) method specifically for PTL reconstruction.
• Utilized a Flying-walking Power Line Inspection Robot (FPLIR) to capture progressive motion sequence datasets of PTLs.
• Implemented spatial compression of unbounded scenes using normal L∞ and employed Integrated Position Encoding (IPE) and Hash Encoding (HE) for sample point data.

Main Results:

• Achieved high-fidelity 3D reconstruction with PSNR = 29, SSIM = 0.871, and LPIPS = 0.087.
• Demonstrated improved integrity and continuity of reconstructed PTLs when using progressive motion sequence images.
• Showcased enhanced efficiency and accuracy in image-based PTL reconstruction compared to existing methods.

Conclusions:

• PL-NeRF effectively overcomes challenges in PTL reconstruction, offering superior performance.
• The integration of PL-NeRF with progressive motion data significantly enhances reconstruction quality.
• This method provides a foundation for automated monitoring and digital engineering of power transmission corridors.