Source object
Here are the results of some objects I scanned. (Note: Some images are shown as left-right mirrors of actual data taken.) The raw data (lots of .BMP's) is here.
Source object
3-D Data
3-D Data - Y-Z projection
Using the Y-Z projection, I was able to easily 'filter' the top of the box from the front face and then independently process each data set. Using JY's FitPlane function (which uses SVD to compute the best-bit plane through a set of points), I found best-fit planes for both faces. Then, using the equations of these planes (and assuming that the source planes were indeed flat and at right angles), I was able to compute some error figures for my data:
RMS error = 0.0248cm, MAX error = 0.0981cm
RMS error = 0.0264cm, MAX error = 0.1033cm
88.3deg
If we assume that the total 'scannable' volume for this setup is approximately 15x15x15cm, these error figures seem quite acceptable (well under 1%). The angular error (ideally, it should be 90deg) is due at least in part to a not-quite-square and not-quite-plane source. Also, my calibration stage is not perfectly square/flat (although it's darn close), and I'm suspicious of the optics on the Quickcam, even after applying the distortion correction.
Source object
3-D Data - Click on plot to see animation
Some Meshes
Click here to get to the VRML 2.0 file. (Note that the initial view-point may not be set correctly in the VRML file, but it's in there!) If you don't already have a VRML plug-in, one source is here.
Source object
3-D Data - X-Z projection (top view)
3-D Data - Click on plot to see animation
When watching the animation, look at how well the lip around the rim of the cup is resolved!
Source object
3-D Data - Click on plot to see animation
When watching the animation, you can see where the view angle is equal to the laser source angle. Note that the scan was only done to approximately the start of the tail.
In this first mesh we are rougly looking 'down' on the dino. The holes on the upper leg and torso do seem to correspond to some missing points that can be observed in the raw 3-D data (above). My mesh-generation code was put together pretty quickly, and no hole-filling or smoothing is being performed.
The final mesh is from a viewpoint in front of the dino. We can see the spatial relationship between the two hands and two feet pretty well here.
Click here to look at the VRML 2.0 file.
