Immersive Reconstruction of Jinhua Intangible Cultural Heritage Space Based on NeRF and Panoramic Image Stitching

Abstract

Accurate three-dimensional reconstruction from limited image data and inconsistent geometry remains a central challenge in cultural-heritage digitization and engineering scene modeling. This problem is also relevant to electromagnetic measurement environments, where sparse-view reconstruction can support virtual testing of antenna sites, propagation scenarios, and complex indoor scenes. To solve this issue, this paper proposes an immersive reconstruction method integrating Neural Radiance Fields and deep panoramic image stitching. The method uses the continuous modeling capability of NeRF for geometric restoration and panoramic stitching to enhance texture integrity. Its implementation includes deep-guided multi-frame stitching, continuous geometric modeling and color fitting through NeRF, and texture-stitching optimization using the PatchMatch algorithm. The process shows strong adaptability under limited image input. Experimental results confirm that the NeRF fusion model maintains stable reconstruction even with only four input frames, reaching an SSIM of 0.65 and a chamfer distance of 0.45. It also achieves good texture continuity in challenging areas, with LPIPS scores not exceeding 0.151. The system provides a strong immersive experience, with a distortion score above 0.9, while maintaining practical efficiency at high resolution. These results provide a useful benchmark for high-fidelity geometric and textural reconstruction in heritage spaces and related electromagnetic scene-digitalization tasks.