THESIS
2007
xiii, 148 leaves : ill. ; 30 cm
Abstract
Many applications for wireless sensor networks, such as object tracking and disaster monitoring, require effective and efficient techniques to detect events of interest in the physical world. Despite the simplicity in implementation, the existing threshold-based approach to event detection suffers from limited expressive power and poor tolerance to missing values and errors in sensor data. In this thesis, we present three novel database approaches to event detection for sensor networks: (i) a pattern-based approach that abstracts events as temporal patterns on individual sensor nodes, (ii) a contour map-based approach that represent events as spatial shapes in the sensor value distribution across nodes, and (iii) a model-based approach that specifies events as regression models over spa...[
Read more ]
Many applications for wireless sensor networks, such as object tracking and disaster monitoring, require effective and efficient techniques to detect events of interest in the physical world. Despite the simplicity in implementation, the existing threshold-based approach to event detection suffers from limited expressive power and poor tolerance to missing values and errors in sensor data. In this thesis, we present three novel database approaches to event detection for sensor networks: (i) a pattern-based approach that abstracts events as temporal patterns on individual sensor nodes, (ii) a contour map-based approach that represent events as spatial shapes in the sensor value distribution across nodes, and (iii) a model-based approach that specifies events as regression models over spatial regions in the sensor network. The key observation of all three approaches is that, because an event generates a particular kind of sensor data distribution, the problem of event detection can be converted into that of matching this distribution. In each of the three proposed approaches, we formalize the event specification and develop efficient matching algorithms. The experimental results on both real-world and synthetic data sets demonstrate the effectiveness and efficiency of our proposed approaches.
Post a Comment