Control Points (CP), also known as Reference Points, are the holy grail of image stitching. They are identified in the overlap area between two images so that the images can be aligned to each other in space.In the old days, panorama photographers would tediously map these points manually, as a set of pixel coordinates on adjacent images. Even before that, they would patiently use scotch tape. Then came automated software.
The area is still the focus of much research, and was the subject of a Google Summer of Code project by Zoran Mesec, mentored by Dr. Herbert Bay, the inventor of the SURF algorythm. The project is still under development and Zoran has added new code to the hugin repository a few weeks ago, just before his exams.
The problem of most current software for automatic generation of CP is that it has troubles with those areas where the two photos have different kind of distortion. This is inevitable with fisheye lenses that are very popular for the production of full spherical panoramas.
A solution to this problem is to compare the photos after remapping them into a conformal projection. Bruno Postle announced recently an implementation of a perl tool that runs the usual automatic generation of CP on temporary stereographic versions of photos. Despite all the apparent overhead, it is not significantly slower than conventional software with the same feature detection algorithm.
For the moment it is a perl script, dependent on some extra CPAN perl modules and ImageMagick. And it runs from the command line. But it won’t take long until Hugin can feed it with the appropriate parameters. I’ve been delaying my next installer snapshot in the hope to be able to add it.