Single image orientation of UAV's imagery using orthogonal projection model

Abstract

A Perspective based image restitution for registering and processing UAV's images is widely known in photogrammetry and computer vision communities. Despite its reliability to compute all necessary parameters to model 3D objects, however it is less suitable to process images which have a very narrow angular field of view. An interested feature on images captured from a very distant imaging object or by a long focal length camera occupies a small portion of the field of view. The perspective model produces unstable results or singular outcomes if the image's field of view is less than 10 degrees. The captured ground features are usually located on small clusters which have very narrow field of view. Therefore, stable registration results are barely achieved. On the other hand, the orthogonal projection model was claimed to give stable results in such situation, particularly for a remote sensing imagery and a non-topographic photogrammetry. This paper demonstrates a feasibility of the orthogonal model based space resection to perform a featured based UAV image registration using surveyed ground control points. Using a minimum of 4 points to reach the field of view of around 6 degrees on each cluster, a five-group of collected clusters is computed. Whilst the perspective model produces erratic results, the orthogonal model gives better solution of exterior orientation parameters in all parts of the image.