Extraction of man-made objects from imagery has been an active research area in photogrammetry and computer vision societies for decades, and feasible solution for simple structures have been reported and showed success on rural scenes. As one of the important map compo1nents of man-made objects, automatic extraction, or less human intervention, of them has great impacts on relieving the bottle neck of data-to-information workflow and generating city models/planning/monitoring applications.
When it comes to reconstruction of urban scenes, matching and reconstruction find difficulties with conventional methods, since linear features which are rich in urban scenes and suitable for describing man-made objects, are not matching and reconstruction entities in conventional point based photogrammetry. Meanwhile, feature-based photogrammetry claiming geometric strength, accuracy in measurements and ability to fuse with other sensor data, has showed potential for effective treatment of linear features. However, notable progress to use linear features has been made in orientation to solve camera calibration and viewing geometry.
This research attempts to fill the gaps which are at their early stages - matching, intersection and grouping of linear features in feature based photogrammetry. The study first focuses on the fullest investigation of geometric constraints made by camera viewing geometry and prior knowledge. Then novel geometry-driven line matching in object space which does not use photometric information and is suitable for multiple image configuration is presented. Moreover, matching takes specifically vertical, horizontal, arbitrary line and orthogonal junction to provide benefits for reconstruction with explicit line information. Proximity, parallelism and perpendicularity are applied to group matched lines, and reconstruction is completed by detecting and adding missing but critical line entities. The output results in three different man-made structures - building complex, C- shaped single building with roof and single building with structures, that show feasible solutions in resolving, complementing and alleviating 1) the limitation of existing matching methods in urban scenes and 2) the limitation in accommodating linear features in matching. Furthermore, output in reconstruction showed that man-made objects can be well-described by lines and junctions, which are matching and reconstruction entities in this study.