Skip to Main Content
Due to their limited expressive capabilities and computational resources, typical mobile devices are still not appropriate for real-time 3D navigation, in particular when widespread building models in the urban environments have a very detailed geometry. In order to better balance the capability of mobile devices and the perception demands of the users in 3D navigation, this paper proposes an automated algorithm to generate structure-preserving abstractions of building models for the purpose of suitable dataset preparation. Building models are segmented into different structural parts, and each structural part is approximated by a convex polyhedron, then the convex parts are combined together to achieve the ultimate abstraction. The robustness and efficiency of the algorithm is proved by extensive experiments.