|
The purpose of data registration is to find the registration
transformation between two point sets. The main application area
of data registration is in reverse engineering and computer
vision. Recently, many researcher focus on this topic and try to
develop an automatic registration system. We use the ICP method
to develop a data registration system and then discuss the
accuracy of registration. The ICP method uses least-squares
estimation to reduce the average distance between the matched
points in the two point sets. It includes two steps: finding the
initial transformation and the final transformation through
iteration. In this thesis, the principal axes are used to find
the initial transformation. The SVD method is then used to find
transformation matrix during each iteration. After registering
two point sets, there are redundant points in the overlapped
area between them. We can remove the redundant points and
resample the two point sets to get a new point set. After data
registration, the accuracy of the six transformation parameters
can be estimated using the proposed uncertainty model. It is
found that the uncertainty of the registration parameters are
inversely proportional to the squared root of the number of
points. This result can be used to predict the minimum number of
points we should use to register two point sets under the
required accuracy so that the computational efforts can be
greatly reduced.
|