Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.652698
Title: Recognition of three dimensional objects using deformable models
Author: Hughes, H. W.
Awarding Body: University of Edinburgh
Current Institution: University of Edinburgh
Date of Award: 1992
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Abstract:
This thesis considers the problem of efficiently identifying and locating instances of classes of three dimensional objects by matching them with a single generic model that represents that entire class of object. Since a member of an object class will normally differ from the prototype that represents the class, the approach used here is to allow the model to stretch, or deform, to fit the object. The input image data may contain one or more objects of unknown identity and location, each of which is assumed to belong to a class of object for which there is a corresponding model. Both objects and models are represented by three dimensional surface data with the object data pre-segmented into surfaces of uniform curvature. The process of deforming and matching the models to the object data is achieved in two stages. In the first stage combinations of object surfaces are formed and a search made for suitable object to model correspondences. Simple constraints are developed to reduce the search space to an acceptable size. When a correspondence is achieved, an initial estimate of the stretch required in the model is made and the model that contains those surfaces is selected for further matching. Because only those models for which there is evidence in the image are selected for further matching, the search space is further reduced. The second stage of the process involves taking those models selected in the first stage and performing a rigorous geometric search for any remaining model to object correspondences. As part of this process, the locations of the objects in the image are predicted and the deformation parameters refined as new correspondences are found. The location and deformation parameters provide further constraints for the geometric search, reducing the search space still more. Recognition is demonstrated with a variety of objects, both synthetic and real, and the results discussed. The use of deformable models in object recognition was found to be a good means by which to represent and match objects from classes showing three types of deformation - scale, stretch and small variation. The model deformation as formulated enabled the identity of the corresponding objects and their parameters of deformation to be determined with accuracy and efficiency.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.652698  DOI: Not available
Share: