A Comparative Assessment of Open-source Building Footprint Data for Indian Heritage Sites: Towards a Shape-preserving 3D mapping

Accurate building footprint extraction is essential for generating 3D models of higher level of detail (LoD2) from satellite imagery. While several open-source datasets and algorithms exist, they are primarily trained on non-Indian buildings and perform well for simple, orthogonal shapes. Indian heritage buildings, however, exhibit diverse architectural styles, layouts, and roof forms, posing challenges for automated footprint extraction. Additionally, existing shape evaluation metrics often prioritize areal similarity and compactness over shape resemblance. This study investigates the suitability of current open-source building footprint data for Indian heritage structures, assessing their performance in terms of semantic and geometric accuracy, with a focus on shape resemblance. By identifying key limitations in coverage, post-processing requirements, and contour approximation, the study underscores the need for an improved approach tailored to India’s architectural diversity. The findings will inform the development of a 3D mapping tool to enhance its footprint extraction for applications in tourism and conservation.