Title: NeRF-Con : Neural Radiance Fields for Automated Construction Progress Monitoring
Authors: Yuntae Jeon, Almo Senja Kulinan, Dai Quoc Tran, Minsoo Park, Seunghee Park
Abstract: The monitoring of construction progress is crucial for ensuring project timelines, budget adherence, and quality control. Traditional methods often involve manual inspection, which is labor-intensive and prone to human error. We introduce NeRF-Con, an innovative approach utilizing Neural Radiance Fields (NeRF) to automate the process of construction progress monitoring. NeRF-Con can infer images that render the construction site with a level of quality comparable to reality by utilizing NeRF, which synthesizes novel views of complex scenes from a sparse set of images. Additionally, by employing a segmentation model, NeRF-Con can compare these rendered images with existing blueprints to gauge the progress of the work. This capability is achieved by training the model using handheld smartphone-captured video. This paper details a method for applying NeRF in real construction sites, encompassing data collection and pre-processing, and compares it with BIM for progress evaluation. In assessing the model's performance, comparisons are made with data from mobile-LiDAR, stand-LiDAR, and BIM. With this research, we suggest potential future studies in applying NeRF models to construction progress monitoring systems.
Keywords: NeRF, 3D Computer Vision, Deep Learning, Segmentation, Construction Progress Monitoring
DOI: doi.org/10.222...
Негізгі бет Paper ID 120 - ISARC 2024 by
Пікірлер