An ISVD-based Euclidian structure from motion for smartphones

Abstract. The development of Mobile Mapping systems over the last decades allowed to quickly collect georeferenced spatial measurements by means of sensors mounted on mobile vehicles. Despite the large number of applications that can potentially take advantage of such systems, because of their cost their use is currently typically limited to certain specialized organizations, companies, and Universities. However, the recent worldwide diffusion of powerful mobile devices typically embedded with GPS, Inertial Navigation System (INS), and imaging sensors is enabling the development of small and compact mobile mapping systems.

More specifically, this paper considers the development of a 3D reconstruction system based on photogrammetry methods for smartphones (or other similar mobile devices). The limited computational resources available in such systems and the users' request for real time reconstructions impose very stringent requirements on the computational burden of the 3D reconstruction procedure.

This work takes advantage of certain recently developed mathematical tools (incremental singular value decomposition) and of photogrammetry techniques (structure from motion, Tomasi–Kanade factorization) to access very computationally efficient Euclidian 3D reconstruction of the scene.

Furthermore, thanks to the presence of instrumentation for localization embedded in the device, the obtained 3D reconstruction can be properly georeferenced.