High fidelity three dimensional reconstruction and interactive visualization of public art sculptures using neural radiance fields

Neural radiance fields (NeRF) offer photorealistic novel-view synthesis, but their use for public art remains sensitive to reflective materials, uncontrolled illumination, viewpoint gaps, and browser-side latency. This study presents a reliability-aware NeRF workflow for sculpture reconstruction and interactive visualization. The scientific contribution is not a new radiance-field representation; it is a visibility-weighted cross-view supervision strategy that uses background, occlusion, and reflection masks to reduce unreliable correspondences, together with a unified evaluation protocol linking reconstruction fidelity, view consistency, and interaction performance. Five indoor and outdoor artworks were captured from 65 to 120 viewpoints. Under a common image split and camera-pose protocol, the proposed configuration achieved a normalized geometric error of 1.20 ± 0.20%, PSNR of 28.46 ± 0.70 dB, SSIM of 0.921 ± 0.010, and multi-view Recall of 96.4%. Exploratory paired tests across the five samples indicated lower geometric error and higher PSNR and SSIM than the two classical pipeline baselines (all p < 0.001), although the small purposive sample limits generalization. The browser prototype reached 27 fps with an initial loading latency of 1.3 s, and the 24-participant evaluation yielded an overall rating of 4.42 ± 0.34. The results indicate that the reliability-aware cross-view strategy improves reconstruction consistency for the evaluated artworks while maintaining browser-side interaction efficiency. Because Instant-NGP, Mip-NeRF 360, and 3D Gaussian Splatting were not executed on the same dataset, the findings are limited to comparisons with the standard NeRF and the two classical reconstruction pipelines.