M.Sc Thesis

M.Sc StudentValdman Liron
SubjectSensors Selection for Complex Event Detection
DepartmentDepartment of Industrial Engineering and Management
Supervisor PROF. Avigdor Gal
Full Thesis text - in Hebrew Full thesis text - Hebrew Version


Complex Event Processing (CEP) is a technology in data processing with different techniques and tools for analyzing data from multiple sources in an efficient ways. In CEP there are various ways to combine events for detecting events in higher knowledge level. This technology is important in the development of applications that deal with voluminous streams of incoming data with the task of finding events and respond to events of interest in real time.

Important sources of events for CEP applications are Wireless Sensor Networks (WSN), which can be used for tracking, battlefield, and health care. A sensor is a type of hardware event producer in event processing. Examples of sensor types are passive infrared (PIR) sensors, GPS location devices, and simple sensors that report aspects of the environment, such as temperature, sound and light. A WSN consists of a large number of sensor nodes, so data of multiple sensors is often combined to reduce costs related to measurements, time and other costs. Furthermore, complex events (such as the detection of a fire) are derived from raw events (e.g., temperature, smoke) that stem from multiple sensors.

In this research, we propose a dynamic CEP’s realization over a sensor network. We propose a Prediction-Update algorithm that consists of two repetitive stages, namely Prediction and Update. The former perform probabilistic event detection by sampling a subset of sensors according to hierarchical selection. The latter uses prediction samples and gets feedback to update the model for the next prediction stage.

Sensors selection has been extensively studied in information fusion. We consider systems with multiple complex events to track simultaneously by hierarchical selection. Further, our approach derives complex events by combining events with historic data. We save the historic data in Histogram Time Series (HTS). Once sensor readings are available, we address the question of uncertain event detection according to those readings and to HTS models. We provide an algorithm for probabilistic event detection of multiple complex events in sensor networks and show its feasibility through a set of empirical evaluations.