|Ph.D Student||Akav Amos|
|Subject||Determination of Relative Orientation with Linear Features|
|Department||Department of Civil and Environmental Engineering||Supervisor||Professor Emeritus Yerach Doytsher|
|Full Thesis text - in Hebrew|
Digital photogrammetric stations had became more and more popular over the last 20 years. One of the main benefits of digital photogrammetry is its capabilities in varies automatic procedures. A major procedure in any photogrammetric task is solving reliable orientation parameters of the scene. Moving from point-based photogrammetry toward feature-based photogrammetry seems as a reasonable step toward achieving fully automatic procedures such as aerial triangulation. Although matching of distinct points is common in the photogrammetric tasks, matched points do not provide useful semantic information. One dimensional and two dimensional features contain more information than distinct points and can be therefore considered as an improved basis for solving orientation parameters and aerial triangulation.
This work focuses on developing and analyzing possible solutions for the classic problem of determining relative orientation parameters using one-dimensional features. Relative orientation determines the geometric dependency between overlapping images within a stereo model. The procedures proposed here are mainly based on using free form 3D curves. Free form curves in image space are usually represented by a sequence of 2D points. Trying to represent such curves in a polynomial or parametric form would yield a more simplified mathematical modeling but at the same time would result in some loss of information due to inherent generalization process that is being involved.
Similarly to other approaches in the computer vision and photogrammetric communities there is a distinction between planar and non planar curves. Regarding planar curves, we start recovering the relative orientation parameters by using free form planar curves while each planar curve adds 3 more parameters to the overall solution. Implementing this mechanism by using the homography matrix leads to a linear transformation followed by a robust solution.
As for non-planar curves, solving the relative orientation parameters is feasible only when there is an epipolar tangency along the curve. Epipolar tangency means that one of the tangents along the curve coincides with the epipolar line.
In order to examine the possibility of using non-planar curves for solving the relative orientation parameters, a tool for reconstructing the spatial curve and its tangent vector was developed. The reversal of the reconstructed tangent vector within a curve with epipolar tangency is presented and used for the final solution.