Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.586895
Title: Energy efficiency and classification accuracy trade-offs in accelerometry-based activity recognition
Author: Wang, Ning
Awarding Body: University of Southampton
Current Institution: University of Southampton
Date of Award: 2012
Availability of Full Text:
Full text unavailable from EThOS. Please contact the current institution’s library for further details.
Abstract:
Driven by growing real-world application 'such as healthcare challenges, accelerometry-based activity recognition has been widely studied as a potential context-aware subsystem for the future pervasive healthcare system. Reliable, accurate recognition and energy efficiency to enable long-term and non-intrusiveness activity monitoring are the key issues for practical use in healthcare applications. A great number of activity recognition systems have been proposed focusing on sensor node design and recognition algorithm development to achieve good classification performance, while the trade-offs between recognition accuracy and energy efficiency has not been investigated in depth. This research investigates this issue by comparing on-node and off-node activity recognition schemes through a practical development. The main contribution of this research is concluded as follows. Firstly, the trade-offs between classification accuracy and energy efficiency is raised as the key issue in sensor based activity recognition system to tackle the real world application challenges. Then this research presents a systematic, empirical design process for optimizing an activity recognition system with respect to the above issue. Such design process involves defining application, designing hardware platform, developing classification recognition algorithm, energy consumption modelling and real system performance evaluation. On-node and off-node classification schemes are the two design philosophies which are compared in this process. This research performs the first step to strike the energy-accuracy trade-offs in body sensor based activity recognition system. The future work should be generalised in two ways. First, different design schemes between the two extreme design philosophies are to be analysed. Then, more classification algorithms should be investigated.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.586895  DOI: Not available
Share: