Image Registration of Kite Aerial Photographs

Abstract

Kite aerial photography (KAP) is an affordable alternative to drone photography, but KAP images are prone to distortion due to camera wobble and field-of-view settings. This thesis attempts to automate stitching of KAP images by identifying and minimizing the factors that prevent successful image registration. Moreover, it provides an experimental analysis on the quality of registration of distorted versus undistorted images. Significant features of a set of images are initially determined through feature detection. These features are assigned an identifying vector called feature descriptors. Similar descriptors among images are matched through a geometric transformation, and are registered as a panorama by laying the similar features onto each other in a single coordinate system. The quality of the registration is determined by the total area of the stitched set of images.