Enhancing NERF Rendering in Architectural Environments Using Spherical Harmonic Functions and NEUS Methods

Conventional neural radiation field (NERF) techniques suffer from color deviations and artifacts or false objects in the reconstruction results when fusing architectural view-points. Due to variations in lighting conditions and inconsistencies in multi-view image data, traditional neural radiation field methods tend to produce inaccurate color renderings and generate false objects. To alleviate these problems, we integrate a ball-harmonic function into the Neus method to represent illumination and color information more efficiently, thereby improving the robustness and accuracy of the reconstruction model. Our method utilizes the spherical harmonic function to efficiently model the lighting environment, which, combined with the geometric representation capability of the Neus method, significantly reduces color bias and eliminates artifacts in the reconstruction results. We conducted experiments on the mip-NERF360 dataset and a customized building dataset to validate our method. The results show that our method performs well in high-quality surface reconstruction, significantly reduces color bias and artifacts, and improves the accuracy and realism of reconstruction results. This study provides effective solutions to common problems in NERF reconstruction and demonstrates significant progress in the field.