Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.499019
Title: A framework for contextual data fusion in body sensor networks
Author: Thiemjarus, Surapa
ISNI:       0000 0001 3516 8944
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2008
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
In recent years, advances in wireless sensor technology have attracted a rapid surge of interest in pervasive sensing. Although extensive measurements of biomechanical and biochemical information are available in most hospitals, the diagnostic and monitoring utility is generally limited to the brief time points and perhaps unrepresentative physiological states such as supine and sedated, or artificially introduced exercise tests. Transient abnormalities, in this case, cannot always be captured. With the emergence of Body Sensor Networks (BSNs), continuous home monitoring of patients in their natural physiological states is now becoming possible. It provides a unique way of context aware sensing with which transient abnormalities and their associated environment factors can be captured. The aim of this thesis is to develop a general framework for context aware sensing. The proposed method consists of an automatic technique for optimal sensor placement and a distributed inferencing approach for balancing bandwidth and power consumption issues. For optimal sensor placement, a multi-objective Bayesian framework for feature selection has been developed. In addition to the discriminative power of each sensor channel, sensor redundancy and communication costs have also been taken into account. For distributed inferencing, the framework is based on probabilistic graphical model for learning model structure. In this thesis, a factor graph is used to facilitate the mapping of the inferencing model onto the physical network so as to ensure optimal resource utilisation. To integrate temporal information, a Spatio-Temporal Self Organising Map (STSOM) has been developed for effective local processing of the sensor signals. The value of the proposed methods has been demonstrated by validation with both synthetic and real datasets. The main contribution of the thesis includes: the development of a multiple objective Bayesian framework for feature selection and an efficient, iterative algorithm for finding the optimal solution set; the introduction of two new criteria for feature evaluation, namely feature redundancy and network complexity, to cater for fault tolerance and minimal resource utilisation for BSNs; the design of a framework for learning and deploying distributed inferencing models for context aware sensing; and the development of STSOMs for incorporating temporal information for efficient activity recognition.
Supervisor: Yang, Guang-Zhong ; Lo, Benny Sponsor: Anandamahidol Foundation ; DTI
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.499019  DOI: Not available
Share: