Use this URL to cite or link to this record in EThOS:
Title: Template reduction of feature point models for rigid objects and application to tracking in microscope images
Author: Boissenin, Manuel
ISNI:       0000 0004 2674 7786
Awarding Body: Sheffield Hallam University
Current Institution: Sheffield Hallam University
Date of Award: 2009
Availability of Full Text:
Access from EThOS:
Access from Institution:
This thesis addresses the problem of tracking rigid objects in video sequences. A novel approach to reducing the template size of shapes is presented. The reduced shape template can be used to enhance the performance of tracking, detection and recognition algorithms. The main idea consists of pre-calculating all possible positions and orientations that a shape can undergo for a given state space. From these states, it is possible to extract a set of points that uniquely and robustly characterises the shape for the considered state space. An algorithm, based on the Hough transform, has been developed to achieve this for discrete shapes, i.e. sets of points, projected in an image when the state space is bounded. An extended discussion on particle filters, that serves as an introduction to the topic, is presented, as well as some generic improvements. The introduction of these improvements allow the data to be better sampled by incorporating additional measurements and knowledge about the velocity of the tracked object. A partial re-initialisation scheme is also presented that enables faster recovery of the system when the object is temporarily occluded. A stencil estimator is introduced to identify the position of an object in an image. Some of its properties are discussed and demonstrated. The estimator can be efficiently evaluated using the bounded Hough transform algorithm. The performance of the stencilled Hough transform can be further enhanced with a methodology that decimates the stencils while maintaining the robustness of the tracker. Performance evaluations have demonstrated the relevance of the approach. Although the methods presented in this thesis could be adapted to full 3-D object motion, motions that maintain the same view of the object in front of a camera are more specifically studied.
Supervisor: Amavasai, Balasundram Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available