Mobile and ubiquitous computing

Pervasive computing embeds wireless communication and computation into material objects, locations and living entities, thus bringing together the physical and the digital into a single information ecology. We explore how such ubiquitous and tangible technologies can be exploited for applications in learning, particularly in their potential to enhance learning experiences through physical, kinaesthetic engagement with digital technologies and digital representations. We also investigate how highly detailed data about the physical and the built environment, captured through the sensing capability of pervasive computing, can support novel learning activities developed around fine-grained observation of our surroundings.

Tuesday, 03 July 2007



Project Leader

George Roussos

This e-mail address is being protected from spam bots, you need JavaScript enabled to view it


Project staff

Michael Zoumboulakis

This e-mail address is being protected from spam bots, you need JavaScript enabled to view it


Project Details

PhD Research, Jan 2007-



Wireless Sensor Networks, Complex Events, Event Programming.


escalation (doc)

escalation (pdf)


Escalation: Event Detection for Sensor Networks


The aim of this research is to design software that programmatically reacts to Complex Events in Wireless Sensor Networks (WSNs). We define complex events as sets of data points that correspond to an interesting change in the underlying process that the WSN monitors.

Nodes within a WSN are power and computationally constrained. Each node runs on a pair of AA batteries and its useful lifetime greatly depends on the workload of individual components such as the radio, the sensors, the program space, etc.

A common goal for application designers for WSNs is to achieve greater network longevity by adjusting two key factors:

  • The balance between computation and communication, and
  • The sleep patterns of the network.

Computation is not as expensive in terms of power consumption as radio communication, so any pre-processing of the data before transmission is likely to have a positive impact on the network lifetime. Similarly, selectively putting nodes to sleep improves the useful lifetime. Many processes monitored by WSNs exhibit the trait of "normality punctuated by panic", meaning that nodes spend a lot of time recording values that are relatively unchanged until suddenly an event occurs.

Our approach

We present a new approach that successfully detects complex events in WSNs. We are using tools from time-series data mining techniques to convert streaming sensor data into a symbolic representation (text). Then we use string distance metrics to compare temporally adjacent sets of characters (that correspond to sets of sensor readings).


The benefits of our approach are:

  • It enables the detection of events that are extremely difficult or even impossible to describe with the use of traditional SQL-like declarative constructs and thresholds.
  • It enables to perform approximate event detection e.g. to detect events that are "like" other events.
  • It offers the possibility for non-parametric event detection. This would be the case where the users do not know in advance the event type they are looking for.
  • It provides for Dynamic Sampling Frequency Management. Since our algorithm continuously compares adjacent sets of characters, it can autonomously make a decision on whether to increase or decrease the sampling frequency based on the relative distance between sets. Furthermore, this decision can be used for network control and sleep schedule optimisation.
  • It offers numerosity reduction; during the conversion from real values to text, data is compressed thus requiring less storage space and costing less power to transmit.

We have implemented our approach for the Telos Sky family of sensor nodes (shown below).