Image Orientation by Fitting Line Segments to Edge Pixels
Sendo Wang
Ph.D. Candidate
Department of Surveying Engineering, National Cheng Kung University
No. 1, University Road, Tainan 70101, CHINA TAIPEI, R.O.C.
E-Mail: sendo@sv.n cku.edu.tw
TEL: +886-6-2370876 ext.839, 835
Yi-Hsing Tseng
Associate Professor
No. 1, University Road, Tainan 70101, CHINA TAIPEI, R.O.C.
E-Mail: tseng@mail.ncku.edu.tw
FAX: +886-6-2375764
Abstract
A new approach to recovering image exterior orientation for photogrammetry by Least-Squares Model-Image Fitting (LSMIF) is proposed in this paper. Instead of solving the space resection of well- distributed control points, the proposed method uses line segments as control lines. The control line segments are projected onto the image based on the approximate image orientation, and fit to the extracted edge pixels by varying the orientation parameters. The image orientation is determined when the projected line segments optimally fit to edge pixels. This paper describes the objective function and the adjustment process of the proposed fitting algorithm. Two examples of determining image orientation for a close-range and an aerial image respectively with the proposed method will be demonstrated.
1.Introduction
Recovering the image orientation refers to restoring the geometric relation between the sensor and the object at the moment of taking photograph, which is the fundamental task of a photogrammetric, remote sensing or computer vision application. The space resection is traditionally applied to solve the orientation for a single image with the measurements of the image coordinates of the control points. This point-based technique has been successfully applied in many analytical or digital photogrammetric applications. However, manual measurements is required to apply the techniques. The higher-level features, such as lines, planes, or object models, are not only easier to be found in the natural environment but also more meaningful than the point features. Therefore, a substantial research on using higher-level features has been carried out both in photogrammetry and computer vision domains in the last two decades (Haralick and Cho, 1984; Mulawa, 1989; Liu et al., 1990; Jaw, 1999; Smith and Park, 2000).
The idea of using straight lines as control elements for photogrammetry was initiated by Mulawa and Mikhail (1988), in which the space resection are re- formulated based on 3D straight-line equations. Since the line equation can be determined by measuring any two points on it, measuring three lines that are not on the same plane will be able to determine the photo’s orientation. Although it requires more measurements than point- based methods, t he measured points on overlapped photos do not have to be identical, which reduces the difficulty of image interpretation and avoids the point-matching problem. This concept has been expanded to using various types of line segments, even the free form curves, by many researches (Buchanan, 1992; Zielinski, 1992; Petsa and Patias , 1994 ; Schwermann, 1994; Forkert, 1996; Schenk, 1999; Zalmanson, 2000) . However, manual measurements of projected line segments on the image are still required.
