File(s) under permanent embargo
Reason: Restricted by author. A copy can be supplied under Section 51(2) of the Australian Copyright Act 1968 by submitting a document delivery request through your library or by emailing firstname.lastname@example.org
Dynamic information-theoretic sensor selection schemes for target tracking applications
thesisposted on 28.02.2017, 01:15 by Razavi Armaghani, Farzaneh
Wireless Sensor Networks (WSNs) provide ad hoc infrastructure for the applications that must operate in remote and harsh environments, e.g. target tracking, wildlife tracking, and environmental monitoring. The primary factors driving such pervasive applications are the flexibility, fault tolerance, high sensing fidelity, low cost, and rapid deployment characteristics of WSNs. Energy efficiency is a critical feature of WSNs because sensor nodes run on batteries. Batteries are generally difficult to recharge after deployment. The principal way of increasing the network lifetime is to minimise the number of active sensors in the region of interest. Target tracking applications, being some of the main applications in WSNs, need continuous location estimations of moving objects. The requirement for high accuracy of estimations poses an additional challenge to WSNs. Accuracy of estimations can be improved by activating more sensors. However, this approach to increasing accuracy can result in higher energy consumption and a shorter network lifetime. Therefore, a reliable and effective sensor selection scheme is necessary to rotate the tracking task between the optimal sets of active sensors, to balance the trade-off between estimation accuracy and network lifetime. This thesis addresses the problem of the accuracy-lifetime trade-off in sensor selection, for three types of target tracking applications: single target tracking, multiple-target tracking, and group target tracking. Particle filtering is a widely used Bayesian estimation method that is capable of solving realistic problems. We propose, develop, implement, and validate predictive sensor selection schemes to find the best set of active sensors heuristically. The proposed schemes take advantage of particle filtering to calculate the sensor information utilities based on the predicted locations of targets. The use of sensor information utilities and sensor energy parameters in the design of selection cost metrics facilitates the formation of best sets of active sensors. In addition, the problem of the accuracy-lifetime trade-off deals with the type of data processing mechanism. The proposed schemes investigate the impact of both local and central data processing on the trade-off. Group target tracking applications demand a compatible clustering algorithm to accurately estimate the groups and their constituent targets. The proposed clustering framework adaptively finds the target groups based on the notation of trajectory mining and graph theory, and is incorporated in the design of a region-based sensor selection scheme for group target tracking. To deal with the accuracy-lifetime trade-off, our sensor selection philosophy is to select a dynamic number of sensors at anytime. This thesis presents implementation and evaluation details of the proposed schemes. Extensive simulations and evaluations have been performed to show the energy efficiency of the proposed schemes in accurate tracking of the single, multiple, or grouped targets. The proposed schemes can be applied in different tracking and monitoring applications, e.g. wildlife tracking, environmental monitoring, bushfire tracking, and traffic management. The application is made feasible by redefining the concept of information utility based on the physical property of interest, e.g. sound and temperature. We believe that the adoption of the proposed schemes would assist any tracking application to dynamically reconfigure the sensor activities, so that network lifetime is prolonged and a high quality of information is attained.