|M.Sc Student||Alexander Manevich|
|Subject||Line Based Photogrametric 3D Modeling|
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Filin Sagi|
|Full Thesis text|
The application of terrestrial photogrammetry for documentation and reconstruction of sites is widely known, due to the ease of data collection, availability of imaging devices, the ability to avoid physical contact with mapped objects, and the existence of well-established photogrammetric documentation methodologies.
Although use of photographic information to generate detailed 3D data may seem an attractive option, it still requires intensive human support in the reconstruction process. To facilitate the modeling, this research studies autonomous means for image based 3D reconstruction.
While image based 3D modeling has been studied extensively in the past, including stereo-vision, multi-view, and keypoint based approach, experiments show that they perform well on simple objects but not when applied on natural scenes. This failure can be explained by their reliance on radiometric information, which may be limited in some cases and the assumptions which they make about surface smoothness and continuity, which may not always apply.
This thesis proposes a geometrically-based modeling using linear information for 3D reconstruction of imaged scenes. Natural scenes contain a rich set of linear information in transition areas of color and intensity and along surface discontinuities. As opposed to most surface reconstruction schemes the thesis shows that usage of linear entities enables exploiting their geometric properties without the need for radiometric based similarity evaluation.
As line based intersection in space is not unique when using image pairs, correspondence cannot be established using this information only. Resolving the correspondence is approached by introduction of a third image, and when using back projection, an efficient association is obtained. The linear features which we consider are not restricted to straight lines only but also to freeform entities. These are approached by simplification of their shape which facilitates their matching. As the research shows, this simplification does not degrade the accuracy. To offer a computationally efficient model the thesis proposes a geometrical sorting of the edges, and hierarchical reconstruction based on edge significance.
The study shows the application of the proposed method on a variety of settings: including manmade objects, bedrock features within archeological sites with strong variation in depth, as well as modeling of vegetated scenes and plants which are characterized by discontinuities and uniform texture. Results show that the application of the method provides an efficient and detailed 3D model.