Esprit "IT for Mobility" 26900
context awareness
results & data
research papers
press articles
seminars & demos
Sensor Data
The source of the systemís input is the set S of basic sensors. In theory, a sensor can be just about anything: a light sensor, a microphone, a global positioning system (GPS), a camera or even a human. In literature, this last kind of sensor is usually referred to as a logical sensor, while the other examples are called physical sensors. The only requirement a sensor should comply with is that it should periodically produce a value that resembles a measurement of a physical phenomenon in its real world context. The formation of such a signal is a digital and unstructured sub-symbolic value. A context description resides usually on a higher level than the outputs of the sensors.

An obvious way of plotting the sensor data is to make a graph of the sensorvalues of time. When a sensor produces a value, this value could be considered as a component of a vector, which has a dimensionality equal to the number of available sensors. This leads to a second method of visualisation (the phase space plot), in which each point in an Euclidean space could represent this vector. Plotting this space requires that the dimension of the space is less than - or equal to three. Thus, some sensors have to be ignored if there are more than three sensors available. 

Example of a Time Series Plot (eight sensorvalues over time):

Example of a Phase Space Plot (vectorplot of three sensorvalues):

Accelerometer Output and Cue (during different activities): <click here>


These are datasets, copied into an Excel document. The information about the current dataset can be found in the spreadsheet itself. They can be found at <> More datasets will be added soon.
  • Tea_k: first demonstrator software for reading of the com port, plotting the sensordata, mapping it to a Kohonen Map, supervising the classification with a Markov Chain, and outputting everything. Programmed in Visual C++.
  • Tea_core: same as Tea_k, but as platform-independent as possible. Has been tested on Windows, DOS and Linux platforms. Programmed in Visual C++.
  • NNS: simulates and generates sensordata with build-in clustering modules such as the Kohonen map, k-means, growing neural gas, and many others. Was developed for comparison of the clustering techniques.
  • Tea Context Reader: to generate datafiles from the com port.
  • Sensor Demo: demonstrates behavior of the sensors on the TEA1 board.
  • Context Demo server: demonstrates the teaching of contexts, with changing audio (mp3-songs) and animations as an application.
A comparison report between the most common clustering techniques available in machine learning: click here. And click here for the comparison plots.
See our papers for the latest experiment results and their descriptions. 

last update: 24/08/2000 by Kristof Van Laerhoven the previous sites can still be found here and here