In ubiquitous computing machine perception is used to infer high-level situations and context information from low-level sensor data. Looking at various example applications this thesis finds that sensors may often precisely detect previously modeled states and objects, however uncertainty is inherently introduced when human intentions are being inferred.
To analyze the effects of machine perception on human-computer interaction this thesis begins with an example application in which it is possible to fully model the important context information. In the example assembly instructions for flat-pack furniture are provided proactively based on the assembly state perceived by sensors embedded in the furniture parts.
To improve interaction with more general context-aware systems a method is proposed which leverages from the fact that people are highly used to and effective at dealing with uncertain information in their everyday life. By simply displaying the momentary system reliability to the user we make the system more observable. In this way the user can adapt to the systems performance easily and use the system appropriately. In several experiments both qualitative and quantitative performance improvements are found when the method is applied.

Stavros Antifakos obtained a M.Sc. (Diplom) in computer science from the Swiss Federal Institute of Technology (ETH Zurich) in 2001. He joined the Perceptual Computing and Computer Vision Group of ETH Zurich as a Research Assistant focusing on ubiquitous and wearable computing. There, Antifakos mainly worked on perceptual computing and human-computer interaction in a EU project, called Smart-Its.