Alignment & Mosaicing of Non-Overlapping Images

There are many existing algorithms dedicated to finding the correct alignment (relative placement) between a set of overlapping images. Moreover, there are even solutions for matching images which are direct continuations of each other and do not include any overlapping sections. These algorithms base themselves on matching overlapping areas using key features or the continuity of the scenery.

The goal of the project is to implement a tool which finds the relative positions in a much more difficult case: There is no overlap, the images are not direct continuations, the distance between them is unknown (but must be limited) and there might be angle variations when capturing the images. In this case there are no common key features and sometimes no direct continuation from one image to the other, and so it is impossible to rely on the algorithms mentioned above.

In this project, we based ourselves on the paper “Alignment and Mosaicing of Non-Overlapping Images” That deals with this particular problem, and we added notable improvements in the capability to align, run-time and the quality of the results.

In the project we implemented every stage from beginning to end without basing ourselves on prior code.