Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.518513
Title: On shape-mediated analysis of spatiotemporal distributions
Author: Thorpe, Daniel
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2010
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
This research describes a new general framework to automate object-based analysis of a spatiotemporal property of a study phenomenon. Raw data is smoothly interpolated to provide a temporal series of surfaces represented as digital images. From each time sample, objects of activity are automatically extracted. These objects may be queried to provide spatial properties and shape descriptive Tchebichef moments. Objects close together from adjacent time samples are associated together to effectively track the phenomenon’s property through space and time. A complete series of tracked objects, from spontaneous appearance to eventual disappearance is termed a Phenomenon Event, and is used principally as a means to analyse how various aspects, such as the raw value or centre of mass of the phenomenon’s property changes over time. Data sources of secondary, possibly explanatory variables, called covariates, are queried and processed in conjunction with the study phenomenon’s data. By correlating properties between covariate objects and phenomenon objects, the nature of any relationship between them may be examined. This general framework was developed using weekly surveillance of Influenza Like Illness (ILI) cases at participating general practitioners (GPs) across France since 1988 as the study phenomenon. Covariate data came in the form of three hourly weather observations at locations across France. These two disparate datasources expose the generality of the framework, as the Influenza data are digital images on disc, and the Covariate data are network accessed raw values stored in a database. This novel approach employs modern shape description techniques from Computer Vision accompanied by geographical methods and traditional statistics. Such a treatment of surveillance data is new to epidemiology, and we hope it will provide a new perspective in the analyis of public health.
Supervisor: Nixon, Mark ; Atkinson, Peter Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.518513  DOI: Not available
Keywords: QA75 Electronic computers. Computer science ; RA0421 Public health. Hygiene. Preventive Medicine
Share: