Covariance-Based Registration Charles V. Stewart The registration problem in computer vision is the problem of finding the transformation that best aligns (registers) a model with a data set or best registers two or more data sets. The goal is to bring the model and the data set or the multiple data sets into the same coordinate system. Solutions to this problem are required in many application domains. In industrial inspection, registration between model and data is necessary for comparing ideal (\nominal") parts with manufactured parts so that defects in the manufacturing process may be identi?ed [32, 33]. In model construction, placing sensor measurements in the same coordinate system is the necessary prerequisite for building complete models rather than models dependent only on individual views. In medicine, registration facilitates treatment monitoring, mixing of sensed data from different modalities, and application of surgical plans developed off-line [10, 13, 15]. Each of these applications requires precise and accurate estimates of the transformation. Department of Computer Science, Rensselaer Polytechnic Institute, Troy, NY 06/24/2002
Covariance-Based Registration
Charles V. Stewart
The registration problem in computer vision is the problem of finding the transformation that best aligns (registers) a model with a data set or best registers two or more data sets. The goal is to bring the model and the data set or the multiple data sets into the same coordinate system. Solutions to this problem are required in many application domains. In industrial inspection, registration between model and data is necessary for comparing ideal (\nominal") parts with manufactured parts so that defects in the manufacturing process may be identi?ed [32, 33]. In model construction, placing sensor measurements in the same coordinate system is the necessary prerequisite for building complete models rather than models dependent only on individual views. In medicine, registration facilitates treatment monitoring, mixing of sensed data from different modalities, and application of surgical plans developed off-line [10, 13, 15]. Each of these applications requires precise and accurate estimates of the transformation.
Department of Computer Science, Rensselaer Polytechnic Institute, Troy, NY
06/24/2002