Project 2
William Pitts and Nicholas Wilson
CSE 455: Computer Vision
Project 2 Artifact
Yosemite
Graf
Test sequence
This one turned out almost identically to the demo image.
Sequence with Kaiden panorama head
In this example, the lighting was very consistent accross all images, resulting in a very good blend between the images. The feature matcher was also able to handle the mirror, which we thought could have caused problems. The many features that the feature detector found helped in making the transformation estimation more accurate than in the other images we took. The only problem was that the two ends don't line up properly when the image is wrapped.
Sequence taken by hand
On this one we had problems with getting the transformations correct. This is due to the fact that not very many features were detected in these images. The fact that there weren't many featres is not very surprising, since many of the images contain mostly featureless walls. The problem is compounded by the fact that a handheld sequence contains rotation transformations, which we did not take into account.