Exploitation of linear features for object reconstruction in digital photogrammetric systems
The availability of digital imagery gives rise to the use of linear features as an additional source of information, or as an alternative to using point features. There are many kinds of linear features. In this thesis, research is focused on straight lines and circles, since they are more frequently encountered in practice than other kinds of linear features. Different parameterization and mathematical modeling for representation of these linear features in both two and three dimensional spaces are developed. Geometric constraints among linear features are also enumerated and functionally described. These constraints provide significant information regarding the exploitation of linear features. There are three main operations involved in the exploitation of linear features for applications in photogrammetry. The first is linear feature extraction or measurement. Before linear features can be extracted from an image, edge pixels designating linear features need to be detected. A concise discussion on edge detection is first presented before different linear feature extraction techniques are reviewed. Some of these techniques are then adopted for use in this research. The second operation involves linear feature matching, in which the correspondence between linear features from overlapping images is established. Previous work related to linear feature matching, especially straight line matching, is reviewed. Different techniques which can be combined for straight line matching are then listed. The last operation deals with photogrammetric image triangulation and object reconstruction. Mathematical equations for various two- and three-dimensional coordinate transformations and photogrammetric conditions needed for both tasks are derived. The developed equations are employed, with the technique of unified least squares adjustment, in several photogrammetric applications, especially image triangulation. An extensive number of experiments, using synthetic and real data, have been performed to test the developed mathematical models and to study the effectiveness of the exploitation of linear features in photogrammetric applications. Results from a representative set of the experiments are shown and analyzed. In the last several years, there has been a great deal of effort attempted at automation of various photogrammetric applications. Many techniques have been developed to reach the ultimate goal where human activity is not required. Such techniques usually work under very restricted conditions. This research follows the philosophy where the human operator's role is to supervise the automated tools provided in the system. A set of automated tools developed for object reconstruction based on linear features is described and their use in a human-supervised approach on a digital photogrammetric workstation, which was used as a development platform, is described.
Mikhail, Purdue University.
Civil engineering|Remote sensing|Earth
Off-Campus Purdue Users:
To access this dissertation, please log in to our