Warning:
This page is
provided for historical and archival purposes
only. While the seminar dates are correct, we offer no
guarantee of informational accuracy or link
validity. Contact information for the speakers, hosts and
seminar committee are certainly out of date.
Shape-based registration is a ubiquitous problem with application to many areas of engineering and science such as medicine, factory automation, and human computer interaction. The goal of shape-based registration is to find a spatial transformation which "best" aligns two data sets corresponding to the same underlying rigid object. Recently, new developments in the area of high technology medicine have provided many opportunities and challenges for the use of registration. In the first part of this talk, I will describe the shape-based registration problem and associated solution methods, and I will present several examples from the medical domain which illustrate the need for robust application of these methods.
Most of the prior research in shape-based registration has been performed off-line using relatively slow solution methods. Within the past few years, sensing technologies have begun to appear which are capable of delivering high quality "range images" at video rates and faster. In order to apply shape-based registration to data from these sensors at high speeds, improvements to existing solution methods were required. In the second part of this talk, I will present a rigid body pose tracker which is capable of estimating the position and orientation of arbitrarily-shaped rigid objects at speeds of roughly 10Hz. I will discuss how this pose tracker could be used in medical and other application areas.
The accuracy which can be achieved by shape-based registration is highly dependent upon the quality of the underlying data. In particular, shape-based registration relies upon sufficient "geometric constraint" between the data sets being registered. Insufficient constraint results in singularities during registration which prevent the accurate determination of a solution. This problem is particularly serious in situations in which data is sparse and data acquisition is expensive. Intra-surgical registration is such an application. Data acquisition during surgery is time consuming, and for certain sensors exposes the patient to radiation. In the last part of the talk, I will describe a technique which can be used to plan for the acquisition of registration data in a manner which maximizes the resulting registration accuracy, while minimizing the quantity of data required. In the context of intra-surgical registration, I will demonstrate how this technique can be used to to significantly reduce the costs associated with performing registration.