Neural rendering-based semantic point cloud retrieval for indoor construction progress monitoring

Abstract

Computer vision has been exploited to retrieve semantic and geometric information for indoor construction progress monitoring. However, existing methods lack the capability to retrieve well coupled semantic and geometric information, which leads to a loss of accuracy and limits the applicability. This study introduces a novel approach called Semantic Reconstruction enabled by Neural Radiance Field (SRecon-NeRF) that extracts highly coupled semantic and geometric information as semantic point cloud. Moreover, a progress estimation strategy is designed to execute progress estimation logic. The evaluation results demonstrate that SRecon-NeRF outperforms the existing semantic-based methods by 24% in accuracy and 75% in speed. It achieves a 36% enhancement in accuracy and an 83.3% boost in speed compared to the geometric-based methods. The utilization of SRecon-NeRF as an information retrieval method in real-world scenarios can improve the practical accuracy, speed, and applicability of CV-based ICPM. Consequently, this can facilitate the widespread digital transformation of ICPM.