|M.Sc Student||Arkhangelsky Igor|
|Subject||Application of Grasp Theory to Registration of 3D Objects|
|Department||Department of Mechanical Engineering||Supervisor||Professor Moshe Shoham|
This investigation deals with 3D registration of free-shape smooth objects.
The success of the registration significantly depends on the quality of the selected reference features. We suggest that there exists a combination of these features, which provides more accurate results. Current research presents a new method of searching for the best reference elements.
Our approach investigates the location of the best sampling features from the point of view of grasp theory. Mathematically, it is possible to consider this problem to be one of trying to grasp and constrain an object with a virtual multi-fingered hand along point contacts. Fingers are modeled as spatial spring-like mechanisms. Utilizing the theory of dual numbers, a novel finger model has been developed, representing each finger as a pair of unit motors (dual vectors).
Starting with a large number of candidate elements, our algorithm, step-by step, filters out less reliable features, singular combinations, etc. The core of the analysis is the idea of principal directions, corresponding to minimum and maximum distances between two sets of reference elements undergoes rigid screw motion. Our research investigates local and global principal directions of two types of reference primitives, homologous points and points on surface patches. The method yields optimum grasp, which corresponds to the best set of reference features.
The method was tested through a full two-stage registration process, including coarse and fine steps. The simulation results show that the proposed approach improves the accuracy of the registration of 3D smooth free-shape objects, both in rotation and translation.