Multi-sensor adaptive context acquisition

My research in wearable computing especially focuses on techniques from machine learning to both train and detect contexts such as the wearer's activity (e.g. running) or environmental situation (e.g. in the park), based on data from simple, small sensors.

To the left is a roadmap of my research into sensing-based wearable computing, click on the numbered links below to download my publications for more detailed information about a specific sub-topic.

1 What shall we teach our pants? This work describes how the user of a wearable device can be taken into the loop to train and detect situations on the spot. The wearable device learns new situations by example, with its user as the teacher. The title refers to the possibility to extend the vision of wearable devices to actual clothing, such as pants.

2 Real-time analysis of Data from Many Sensors with Neural Networks focuses on the strengths of having many sensors distributed over the body, instead of few strong sensors (like microphones and video cameras). The main topic of the paper is the finding that in this view, artificial neural networks are a natural extension to this methodology, also relying on many simple elements.

3 Multi-Sensor Context-Aware Clothing then asks what additional value a multitude of body-worn sensors can actually give by looking at one specific hardware prototype using 30 motion sensors. Three dimensional plots are tried to examine the worth of the sensors, and the contexts to be recognised by the sensors.

4 Context Awareness in Systems with Limited Resources takes a very pragmatic view, concentrating on adaptive techniques to train and recognize contexts with only a basic microcontroller to work with. A statistical method is proposed that is able to flexibly realize context awareness, without needing too many resources.

Last updated by Kristof Van Laerhoven, 18/01/2003 17:01:52 -0000