Making Context Work

the foundation for context aware applications

Defining Context

One of the central problems in building context-aware applications is defining what context is relevant and how can it be sensed/classified/modeled. Without an understanding of how context and application interact nothing else can be done. In addition, some types of context are relatively easy to determine (such as location) and some may be quite hard (such as "talking to my friend Bob").

Sensing Context

The selection of effective sensors and signal processing techniques is a topic worthy of at least two graduate level courses alone. One of the most important decisions is in how to measure the context of interest; can the context be extracted from low-bandwidth sensors using relatively simple techniques, or are more expensive and complex approaches required?

For example, if the context of interest is "talking to Bob," a number of sensing strategies are possible. Cameras and expensive computer vision techniques may be employed, or (if the situation is appropriate) a simple low-bandwidth tag and tag-reader might solve the problem.

Classification and Modeling

The result of measurement is a signal, and this signal requires interpretation. The method of interpretation depends on the type of signal (low or high bandwidth, the dimensionality of the signal, type and quantity of noise, etc.) and the context of interest. If the sensor is a tag reader, the interpretation may be as simple as a table lookup. In the case of a camera, most contexts of interest (such as "talking to Bob") will require expensive and sophisticated computer vision techniques.