A line photogrammetry algorithm for 3D rectilinear object reconstruction
Master Thesis
1998
Permanent link to this Item
Authors
Journal Title
Link to Journal
Journal ISSN
Volume Title
Publisher
Publisher
University of Cape Town
License
Series
Abstract
This thesis introduces an alternative formulation for line photogrammetry. The aim was to develop and test a method of computing the position and orientation of a straight line in space using two or more oriented images of that line. The algorithm presented is intended for object reconstruction and is motivated by the need to reconstruct man-made objects in urban areas, such as buildings and the industrial inspection arena. The method aims to obtain a best-fit line through a "pencil of planes". The reconstructed 3D line is defined by two points as opposed to the conventional representation, which uses a point and a direction vector. The approach to this problem involves the calculation of a projection plane for each image containing the perspective centre and two transformed line-point observations in the image. A least squares adjustment involves fitting a straight line as near as possible to the projection planes from all images simultaneously. The adjusted line is referred to as a best-fitting line through a "pencil of planes" (POP). In this project, a mathematical model was formulated for the application of this concept. This algorithm was coded and tested on two cases. A set of scanned aerial images of a residential area with a scale of 1: 5000 provided the primary test case. Lines delineating three roofs visible in the aerial images were reconstructed using the POP method and compared with ground truth data. The lines reconstructed using the POP method were compared to those reconstructed using an existing method of line photogrammetry, proposed by Mulawa (1988). The second test was based on a set of close-range images captured using a small-format digital camera. Lines delineating the bars of a metal frame generally used as a precise control field for camera calibration, were reconstructed. In both test cases, X² tests were applied, and the standard deviations calculated. In the aerial case, standard deviations obtained were generally in the region of about 5cm. The ground resolution of the images was 7.Scm. In the close-range case the ground resolution was approximately 1.3mm, and the standard deviations obtained were generally of the order of 0.7mm. Of the lines computed, 84% of the adjustments passed the X² test. The results obtained confirmed that the POP algorithm is a practicable means of adjusting observations to obtain best-fitting 3D lines using observations made in a set of oriented images.
Description
Keywords
Reference:
Hill, J. 1998. A line photogrammetry algorithm for 3D rectilinear object reconstruction. University of Cape Town.