Bending the doming effect in structure from motion reconstructions through bundle adjustment

Abstract

Structure from Motion techniques provides low-cost and flexible methods that can be adopted in arial surveying to collect topographic data with accurate results. Nevertheless, the so-called "doming effect", due to unfortunate acquisition conditions or unreliable modeling of radial distortion, has been recognized as a critical issue that disrupts the quality of the attained 3D reconstruction. In this paper we propose a novel method, that works effectively in the presence of a nearly flat soil, to tackle a posteriori the doming effect: an automatic ground detection method is used to capture the doming deformation flawing the reconstruction, which in turn is wrapped to the correct geometry by iteratively enforcing a planarity constraint through a Bundle Adjustment framework. Experiments on real word datasets demonstrate promising results.