3D Reconstruction of Construction Sites Using Neural Radiance Fields for Progress Monitoring

Effective progress monitoring on construction sites is essential for project success in that it helps mitigate risks associated with construction delays. Unmanned Aerial Vehicles (UAVs) are widely used to collect spatial data from construction sites on a large scale without requiring humans to visit areas. 3D point clouds are then generated using photogrammetry or laser scanning techniques for site modeling while performing field inspection and earth-volume estimation tasks. However, existing 3D reconstruction technologies have disadvantages in depicting amorphous or reflective objects. This paper introduces a recent approach utilizing Neural Radiance Fields (NeRF) for 3D reconstruction of construction sites from UAV-captured data. The reconstructed models were evaluated against ground truth data using nearest neighbor distance, and a quantitative evaluation was performed using the Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) score. The findings demonstrated that NeRF can effectively synthesize photorealistic views of complex scenes, improving the fidelity and utility of 3D models for construction site monitoring. The practical applications can be extended to semantic segmentation and volume estimation, further contributing to its value in automatically quantifying construction progress.