True 3D-information from digital aerial images—Edge detection and adjustment from an oblique viewing multi-camera system

Abstract

Automated processes for building reconstruction including roof- and façade structures with accuracies better than 0.5 m have not been developed yet. This paper introduces a conceptual approach to derive 3D building edges automatically from oblique aerial imagery. A sub-pixel accurate version of the Canny algorithm is used for detection and segmentation of line segments found on seven oblique true orthomosaics (TOM) mapping a scene from different cardinal directions. With respect to the properties of oblique photogrammetric products and the characteristics of their derivation, all detected line segments are weighted based on their individual camera-target path. An adjustment approach is implemented, which adjusts redundant line segments that are found in several oblique TOM. A validation of the adjustment is done by using tacheometrically measured reference points. On average the approach yields in accuracies of 30 to 35 cm in X, Y and Z.