Immovable Cultural Relics Preservation Through 3D Reconstruction Using NeRF
Zhiqing Lai, Mingchang Lian, Songhang Chen, Guoqing Chen, Xiaoling Lai · Frontiers in artificial intelligence and applications · 2024
Urban and rural development poses significant threats to immovable cultural relics, leading to their degradation and potential extinction. To address this issue, 3D reconstruction has emerged as an effective preservation method. However, traditional radar-based techniques require close scanning and extensive manual intervention, limiting their efficiency. This paper explores the application of neural radiance fields (NeRF) for visually reconstructing immovable cultural relics. NeRF employs a fully connected non-convolutional neural network to generate high-quality 3D models from 2D images. The study discusses NeRF’s sampling strategy, positional encoding, field architecture, and rendering process. Experimental investigations evaluate the impact of Fourier parameters on reconstruction quality. A case study reconstructing Tianhou Temple using NeRF is presented, comparing it with ContextCapture software. All results demonstrate that NeRF offers an efficient, and cost-saving solution for reconstructing and preserving immovable cultural relics, providing accurate and realistic models with superior structural integrity.