Orthophoto generation with gaussian splatting: mitigating reflective surface artifacts in imagery from low-cost sensors

Abstract. Orthophotos offer a distortion-free and high-resolution depiction of architectural and mechanical elements, serving critical functions in reverse engineering, defect detection, and condition assessment. Recent advancements in low-cost sensors have made them increasingly popular for orthophoto generation due to their affordability and ease of use. However, reflective surfaces pose significant challenges in traditional photogrammetric workflows, leading to inaccuracies in feature matching and 3D reconstruction. This paper investigates the integration of Gaussian splatting into the orthophoto generation process as a solution to address these challenges. Gaussian splatting is particularly effective in handling irregular and sparse data, making it suitable for scenarios involving reflective surfaces. In this study, datasets containing reflective surfaces, such as metallic elements and urban environments, were used to evaluate the performance of Gaussian splatting compared to traditional photogrammetry workflow. Our findings indicate that Gaussian splatting effectively reduces artifacts caused by reflections while preserving geometric accuracy and critical detail in non-reflective areas. Additionally, this approach proves to be computationally efficient, making it ideal for low-cost sensor applications. Although limitations remain, such as smoothing effects that may reduce fine detail, the proposed methodology shows promise for improving orthophoto quality in complex environments.