The accuracy of the camera calibration is important for the overall look and accuracy of the final image. To start, I estimated all of the camera parameters. The goal was to get the projected textures for the model working first, then get the camera calibration more accurate to see the improvements. I never finished the projection in order to change the parameters. The camera in the following image is the red ball, and the blue vector is the direction of the camera. The next image is a view from one of the cameras.
A simple model of the front right side of the computer science building was created. I wanted to create a full model but figured it was more important to get the other parts working first.
To get a textured model first, I manual mapped U, V coordinates from 3 different photos. It looks relatively well, as can be seen from the overview image above, but I ran into some weird "bending" issues with the textures. The "bending" was a shift in angle when being mapped, usually resulting in a diagonal line, as seen from the next few images.
I found an example from David Yu of SGI that projected an image on a box. I took that example and applied it to my model. A preview of the projection can be seen below.
Unfortunately, I ran into a lot of problems with the matrix transforms and frustums. I finally got it to work, kinda, but too late, so I couldn't perfect it. It ended up stretching along a wall when, I think, it shouldn't have. An example can be seen from below.
This was just a goal, and never implemented. Once the projection was accomplished, I was going to correlate the edges of the image to the edges of the model.
