Description:
At this step we are supposed to have a lot of views covering the whole object, paying special attention to each detail and making sure that the views, when put together, do not leave any uncovered regions of the object. My attempt was to acquire 26 views of the statue, but I ended up using only 11 of them, for reasons described below.
Using a program written here at Caltech (3dpalign) we can compute a transformation for each view (a rotation and translation) such that, when applied to each cloud of points, the points get translated in the same frame of reference and they get to be aligned in the right order and position.
The hard part of this step is to come up with an initial guess, that is, to align individual pairs of two views. This seems to be easy and it really is, but when all the views are put together with the repective initial guesses, thigns may fail to align. When you work with 26 views, you'll find out it is not very exciting to start the alignment process from the beginning, with a new guess.
Code:
Students were not supposed to write code at this point. We used a provided alignment tool 3dpalign.
Results:
Comments:
I started with 26 views and the alignment failed because of the initial guess. Of course, I tried to change the initial guess but no progress was observed. Then I selected 18 views out of the initial 26, but again, the views refused to align. Then I cut down the number of views to 10, they aligned very well but the views were not enough for the next step. A top view was needed, so I added that and 11 seems to be the magic number for my statue.
Update: Actually I had to come back to this step, from the next step (volume carving and marching cubes), as the alignment did not work quite well. I had to also get back to the lab and acquire a new data set, this time with 14 images. This proved to work fairly well, but it is far from being perfect. Definitely, the alignment tool needs some more work, as I experienced problems with views that needed a little bit of sliding motion in order to be aligned.
In general, it is not a good ideea to use a lot of views, even if the alignment is successful, because at the next step, volume carving, the processing time would be increased considerably.