One nice feature of this location is that as the camera pans, we see objects both close and far away, allowing comparison of the effectiveness of the algorithms on objects at various distances. Since the motion of nearby objects is greater from picture to picture, these pictures often have slight ghosting. For the most part though, the books appear to align well, but the human subjects moved slightly from one picture to the next. We attempted to place ourselves in the pictures more than once, with ok results. |
stitch-assist mode. While the stitch-assist allows for alignment along the horizontal axis, it is somewhat harder to keep the camera vertically aligned (at least, when most of the lines you're dealing with in the image are vertical) Due to the smaller control over the translation from image to image, there is a lot of ghosting in this panorama. It was also very difficult to keep the camera rotating about the same axis, which probably has the greatest effect on the errors in the panorama. Also, as I got to the end of the sequence, I improved at taking the pictures, so the quality of the panorama also improves. |
a focal length of 674.82258 pixels, and radial distortion coefficients of -.21528 and .30098. LucasKanade was run with an estimated translation of 200,0 and 4 pyramid levels with 3 iterations each Finally, the photos were blended with a blending window of 200 pixels. |