Project topic posted by Pat Morin

Finding Corners in 3-D Scanner Data

3-dimensional scanners (laser range finders and coordinate measuring machines) can measure an object by taking a sample of points on its surface. These points are usually then fed into some reconstruction software that attempts reconstructs the surface of the object from this point data. For the most part, this software works well, particularly when the object is smooth (like a ball). However, where these algorithms tend to break down is in maintaining sharp corners. For example, the two images below show a perfect cube and a reconstruction of a cube from a sample of 6000 points on the surface of a perfect cube. Notice that in the second image, the corners are not quite as sharp.

The goal of this project is to develop efficient and accurate algorithms to detect corners in 3d-point data and perform reconstructions that preserve these corners. Efficiency is as important as accuracy here because the resolution of 3d scanners is becoming such that the number of sample points is huge.