Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.724696
Title: Online learning of personalised human activity recognition models from user-provided annotations
Author: Miu, Tudor Alin
ISNI:       0000 0004 6425 638X
Awarding Body: Newcastle University
Current Institution: University of Newcastle upon Tyne
Date of Award: 2017
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
In Human Activity Recognition (HAR), supervised and semi-supervised training are important tools for devising parametric activity models. For the best modelling performance, large amounts of annotated personalised sample data are typically required. Annotating often represents the bottleneck in the overall modelling process as it usually involves retrospective analysis of experimental ground truth, like video footage. These approaches typically neglect that prospective users of HAR systems are themselves key sources of ground truth for their own activities. This research therefore involves the users of HAR monitors in the annotation process. The process relies solely on users' short term memory and engages with them to parsimoniously provide annotations for their own activities as they unfold. E ects of user input are optimised by using Online Active Learning (OAL) to identify the most critical annotations which are expected to lead to highly optimal HAR model performance gains. Personalised HAR models are trained from user-provided annotations as part of the evaluation, focusing mainly on objective model accuracy. The OAL approach is contrasted with Random Selection (RS) { a naive method which makes uninformed annotation requests. A range of simulation-based annotation scenarios demonstrate that using OAL brings bene ts in terms of HAR model performance over RS. Additionally, a mobile application is implemented and deployed in a naturalistic context to collect annotations from a panel of human participants. The deployment is proof that the method can truly run in online mode and it also shows that considerable HAR model performance gains can be registered even under realistic conditions. The ndings from this research point to the conclusion that online learning from userprovided annotations is a valid solution to the problem of constructing personalised HAR models.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.724696  DOI: Not available
Share: