ICP should have a "too far way" threshold
This issue was created automatically from an original Mantis Issue. Further discussion may take place here.
The current implementation of ICP requires ALL of the source points to match to the target. This is ok if one set is a strict subset of the other, because you can simply match the subset to the entire set. However, if the polydata which contains the subset also has many other points, the transform gets pulled to the center of the entire dataset rather than staying near the original location. The idea is that some points are "too far" to be considered in the transform at all, so they do not affect the cost function. This is always a problem when matching a model to a LIDAR scan.