VRML representation of objects reconstructed from laser scans

By Adam Fields

for CS6998 - 3D Computer Modeling
  1. Bunny
    This first object was fairly easy to reconstruct from the range scans. There were about 10 scans, which overlapped nicely and, more importantly, were all aligned for merging. From those scans, I was able to recreate a mostly watertight mesh representation for building. The model still had some holes, which were fixed in quickslice. After slicing, there were about 10 or 15 slices that required touching up, but the final bunny (sitting on my desk) is proof that those changes were minor. This mesh was made at level 3.
  2. Dragon Head
    For the second object, I chose something difficult. My first choice, a 6" Tick action figure, did not scan well due to reflections from the surface and the fact that the figure was too tall for the scanner. The second object that I chose, a green plastic dragon head (actually a replacement Rollerblade brake pad), worked better, but had a lot of surface detail that was not fully captured. My biggest problem in performing the reconstruction was in getting the mesh pieces aligned correctly. The closest average alignment I could get was .8-.9, much higher than the recommended .3. The first set of scans, four corners, was not sufficient to reconstruct a watertight mesh, but with the addition of four more corner scans (offset by 15 degrees) and a top scan, I was able to reconstruct the model shown here. I did not use all of the scans, but fit in as many as I could. I think I ended up using seven of the total nine. Even with "by hand" adjustment, the alignment was still off, but I managed to get a pretty good approximation of the original object. The model still has many holes and what appear to be "dead" areas where one area of the mesh didn't overlap in all three dimensions and so sticks out over another mesh without connecting. This mesh was made at resolution 2, in an attempt to get better overlap, but I think all I did was take more time. I think the problems with this reconstruction were as much with the alignment interface as with the scans themselves. It was very difficult to align the mesh pieces in three dimensions working with a mouse and no orthogonal views. It was hard to tell in which direction the pieces were moving. All things considered, however, I'm pleased with the results.
Back
Send mail to Adam