Abstract: There are many situations in which we wish to overlap images
form a larger image. In the most general situation this involves solving a
complex minimization equation to recover bot the camera geometry and the
complete 3D relationships between the images. However this is often
unnecessary because we can assume that the images are all level and follow
one another to form a panorama.
In this seminar, which is a rerun of a paper I presented at ACCV2000,
will describe a rapid, low-cost solution to this problem, aimed at creating
convincing 360¡ panoramas from relatively few images.