Human supervised tools for digital photogrammetric systems

Digital photogrammetric systems, such as softcopy stereo workstations, are beginning to replace analytical plotters for the extraction of cartographic data. They should not be viewed as mere upgrades, however; their potential lies in increased automation and providing added value by expanding the domain of applications and products beyond regular mapping. This dissertation explores the development of automated tools for model setup and mensuration on such systems, specifically pass point selection and matching for use in aerotriangulation, and refinement of automatically produced digital elevation models (DEM\u27s). The goal is to maximize the potential for success of the automated tools by maintaining operator involvement for high-level tasks. Digital image matching and automated blunder detection techniques, which are crucial for automating photogrammetric processes, are reviewed. An algorithm for automatically selecting and matching conjugate pass points is developed. The human operator selects areas of high information content, and an interest operator is used to choose candidate pass points. An adaptive area-based hierarchical correlation algorithm with optional least squares matching refinement is employed to locate the conjugate points. A subset of these points is selected and used in a relative orientation employing automated blunder detection to remove any previously undetected mismatches. Testing of a wide variety of interest operators, matching variations, and blunder detection methods is performed to refine the algorithm. The high quality and robustness of the final version is ascertained by tests on an extensive set of real image pairs and triplets. Orthophotos made from the left and right images of a stereopair will exhibit differences at locations where the DEM is in error. These mismatches are determined with adaptive area-based matching and converted to elevation corrections, which are applied to the DEM to improve its quality. Since the procedure is inexact, it is by necessity iterative. The algorithm is tested on a number of real image pairs, and its performance is evaluated by comparison with manually measured DEM\u27s. In most cases, a reduction in the magnitude of error in the DEM\u27s, and thus the time required for manual editing, is noted