CAREER: Methods for Nonrigid Motion Analysis

University Of Delaware, Newark DE

Investigators

Abstract

This is the first year funding of a four year continuing award. This proposal outlines an extended research program to investigate a hierarchy of nonrigid motion classes, along with their identification and analysis. The educational component of this plan involves graduate and undergraduate course development, efforts for a synergy between research and teaching activities and development of active learning strategies in the proposed courses. Invariant measures from differential geometry are mainly used for our purpose of nonrigid motion analysis. We propose algorithms for general nonrigid motion analysis, based on new distance measures using discriminant, unit?normal and Gaussian curvature variations of surfaces. The proposed system consists of a hierarchy of nonrigid motion models arranged in a simple?to?complex fashion. Our emphasis is not only in designing different classes of nonrigid motions and proposing different techniques to extract motion parameters and point correspondences, but also in applying these algorithms to extensive sets of real data involving face and cloud motion, in particular, hurricane motion. We feel that this is a good testbed for designing, testing, validating and evaluating our system, as face and clouds involve a wide variety of nonrigid motions. This proposal seeks to address three major scientific issues: (i) How to efficiently estimate point correspondences and motion parameters during a general nonrigid motion? (ii) What motion parameters are to be used for an accurate model fit of a given unknown nonrigid motion? (iii) How to classify nonrigid motion? The strengths of the proposed research are threefold. First, the use of invariant differential geometric parameters for nonrigid motion analysis allow the ability to be independent of coordinate system and parameterization, especially since our optimization functions are based on before motion frame's principal coordinate system. The second strength of our approach lies in its ability to interrelate the motion parameters of all the nonrigid motion classes in the hierarchy. The third strength is the ability to estimate point correspondences, nonrigid motion parameters and nonrigid motion type through a single optimization procedure. The educational plan involves integration of research activities in teaching and efforts for undergraduate research involvement. Proposed new courses include graduate level courses in computer vision, video computing and seminar course in analysis and visualization of formable bodies.