Efficient strategies for reducing the size of search space in three-dimensional object recognition and pose estimation
Abstract
In this thesis we show how large efficiencies can be achieved in model based 3-D vision by combining the notions of discrete relaxation and bipartite matching. The computational approach we present is empirically interesting and capable of pruning large segments of search space--an indispensable step when the number of objects in the model library is large and when recognition of complex objects with a large number of surfaces is called for. We use bipartite matching for quick wholesale rejection of inapplicable models. We also use bipartite matching for implementing one of the key steps of discrete relaxation: the determination of compatibility of a scene surface with a potential model surface taking into account relational considerations. While we are able to provide the time complexity function associated with those aspects of the procedure that are implemented via bipartie matching, we are not able to do so for the interative elements of the discrete relaxation computations. In defense of our claim regarding computational efficiencies of the method presented here, all we can say is that our algorithms do not take more than a couple of iterations even for objects with more than 30 surfaces. Methods for range data collection and segmentation of such data are essential elements of our 3D robot vision system. In this dissertation, we will also discuss a new type of a structured-light scanner; we call it the Cross Scanning Structured Light Scanner because two crossed laser beams are used for scanning a scene. We will show how surface features used in our recognition scheme can be extracted directly from the data produced by this scanning system.
Degree
Ph.D.
Advisors
Kak, Purdue University.
Subject Area
Electrical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.