Automation of photogrammetric operations using advanced digital image matching techniques

Automation of photogrammetric operations using digital imagery is the principal objective of this thesis. A shape-based image matching procedure that uses edges as primitives to solve the stereo matching problem in photogrammetry is described. This procedure follows a sequence of operations based on the following steps: edge detection, line segmentation and shape matching.;The procedure has been developed to deal with complex, large-scale, aerial images which present several discontinuities that often imply difficulties in image matching. However, because of the image complexity and the restrictive nature of this shape-based matching procedure, only a few pairs of shapes can be matched from the large number of features usually found in this type of imagery.;In order to improve the procedure, three extensions have been proposed. The first extension is to apply these procedural steps to multispectral or color image data (when available). It has been empirically demonstrated that more pairs of shapes can be matched when applying the procedure to each independent color channel or spectral band and finally combining the results. The second extension is to add a geometric grouping operation to the procedure. This allows the connection of short line segments, broken during the edge detection and line segmentation processes, to form stronger features for matching. The third extension is to develop an algorithm to match straight lines and other shapes that could not be matched by the procedure alone. The straight lines are then used as primitives and the transformation coefficients of the already matched shapes as constraints for the extension of the matching procedure. This extension has been designed to be fast and robust, and to provide a larger number of matched features.;Furthermore, a method for measuring the disparity between the matched features and their accuracy has been developed. A disparity accuracy between 1.0 and 2.0 pixels can be achieved with the automatic matching procedure in comparison to manual observations. Finally, some recommendations for extending this matching procedure to the automatic extraction of features for Geographic Information System (GIS) databases are given.