CSE 455 Computer Vision, Winter 2006

Project Discussion:
The project overall works well with the images I took with the Kaidon Head. I used the same initial vector (250, 0) in calculating Lucas Kanade and the pictures already align very well. The only problem is that the last few images that connect the two end of the panaramic image together has a slightly different motion, so I have to tweak the vector a bit to get the alignment correct. One other problem is the sudden change of lighting on the mountain. This affects the Lucas Kanade once in a while.

However, the images that I took without the Kaidon head doesn't works very well. One thing is that, the overlapping area between any two adjacent images are quite different, ranging from 100 pixels to 400 pixels. Using just one initial vector doesn't work in the case. Moreover, the images are somewhat dim and have a blue shading over it. These two factors make the already difficult to match snowy scene more difficult. Guessing the initial vector was rather difficult in this case. It takes many try to get the images to somewhat align correctly. I had to use a different initial vector for almost each pair of images in order to align them correctly. The camera was also not rotating exactly around its optical center. That leads to some images be able to match only part of the scene.