Gaussian Splitting Photography Strategy for Light Reflection and Repetition Structure in Corridors

Recently, Gaussian Splatting techniques have attracted considerable attention in the field of 3D reconstruction. However, they exhibit limitations in the visual quality of reconstructed models, particularly in narrow environments such as corridors. One of the main reasons for this limitation is that the photographing path of the input image is not accurately defined. Existing studies have experimentally applied various camera paths. However, the effects of environmental factors such as strong light sources and repetitive patterns on reconstruction performance have not been fully considered. In this study, we propose a photographing path that enables stable 3D reconstruction even in spaces where strong light reflection or repetitive structures exist. The proposed path is designed based on the actual corridor environment, and includes lighting directions at the time of photographing, securing parallax between cameras, and strategies for distributing repetitive patterns to improve the visual quality of Gaussian Splatting. As a result of our experiment, this path secures the robustness of light reflection and repetition structure compared to the existing path, with the highest visual quality of 0.799 in SSIM, 24.170 in PSNR, and 0.422 in LPIPS.