University of Twente Student Theses


Statistically-based event detection using Wireless Sensor Networks

Ogbagabriel, Tiblez (2012) Statistically-based event detection using Wireless Sensor Networks.

[img] PDF
Abstract:Widespread spatial and temporal monitoring of the environment is required for drawing accurate scientific conclusions and for providing reliable forecasting. Traditional environmental measurement systems use expensive sensing stations which makes it difficult to deploy a large number of sensing units. Wireless Network Sensors (WSNs) emerged as a solution to such issues. WSNs can be deployed in great numbers to observe the environment with high temporal and spatial resolution. On the other hand, WSNs are constrained by computational, storage and communication limitations. Event detection is an important application in the practical deployment of wireless sensor networks (WSNs). One of the key challenges in detecting events using WSNs is how to detect it accurately and fast (real time) in an online and decentralized way. In order to assure reliable detection of interesting events using WSNs, we need to develop an accurate outlier detection technique while paying attention to the computational, storage and communication limitations in WSNs. As a result there is a need of developing analytical techniques for outlier and event detection which can fit to the resource constraint nature of WSNs. This research takes advantage of the spatial and temporal correlations that exist between sensor data in order to ensure reliable detection of events using WSNs, while maintaining the resource consumption of the WSN to a minimum. The analysis was done based on data from a real world deployment on a high mountain pass (the Grand St. Bernard pass) in Switzerland. The spatial and temporal correlation in sensor data is exploited by using statistical approaches. This research provides domain specific definition for outliers based on the guidelines of World Meteorological Organization (WMO). Strategies for defining events and distinguishing them from errors are provided based on temporal and spatial correlations. This research proposed detecting obvious outliers using plausible value and minimum variability checks and investigated the use of model independent and computationally simple outlier detection techniques for detecting unclear outliers. The outlier detection accuracy of the proposed techniques was evaluated using detection rate (DR) and false positive rate (FPR) based on the results of three labelling techniques (running average-based, Mahalanobis distance-based and density-based) by Zhang (2010). Event detection accuracy was evaluated by relabeling these labelled datasets so that they can be used to evaluate the event detection. This research examined the use of geostatistical analysis for modelling spatial correlation using the variogram averaging method of Sterk & Stein (1997) which is based on the assumption that a constant correlation exists over time. The prediction accuracy of the spatial correlation model was evaluated using leave-one-out cross validation technique and the assumption of constant spatial correlation over time in which the variogram averaging method of Sterk & Stein (1997) is based in is verified. Keywords: Outlier detection, Event detection, Wireless sensor networks, Time series analysis, Geostatistics, Temporal correlation, Spatial correlation.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page