Mapping field photographs to textured surface meshes directly on mobile devices

Abstract

The mapping of photographs to surface geometry is an important procedure for many applications within the geosciences. This paper proposes an interactive framework for feature-based image-to-geometry mapping that works directly on mobile devices, under challenging imaging conditions and with limited available hardware performance. The framework makes use of openly available digital elevation models (DEMs) together with mobile position-and-orientation sensor data. It integrates calculation heuristics for result evaluation and feedback, synthesising available knowledge in current registration literature. The approach is assessed on two image datasets captured on separate occasions. Their interpretations are mapped to one textured lidar surface model and the projection accuracy is qualitatively assessed. The experiments show a significant accuracy improvement in photograph registration results, as well as the faithful mapping of image interpretations on the underlying surface geometry. This semi-automatic, user-guided, interactive approach is superior to comparable fully automatic registration methods.