Abstract: To address the challenges of 3D reconstruction in low-illuminance indoor environments of large public buildings, this paper proposes a rapid reconstruction framework integrating panoramic vision and 3D Gaussian Splatting (3DGS). Panoramic video streams are used for efficient local-space data acquisition, while high-sensitivity images are processed with COLMAP for camera pose estimation. The Brush engine is introduced for radiance field training and interactive denoising, which effectively suppresses noise in dark regions and compensates for geometric holes. Finally, the reconstructed results are exported in a lightweight format via SuperSplat. Experimental results show that the proposed method can faithfully reproduce the texture of highly reflective materials, improve processing efficiency by several times compared with conventional methods, and support smooth web-based roaming. The framework thus provides a highly practical technical paradigm for the digital operation and maintenance of existing buildings.