Information Technology Reference
In-Depth Information
interpersonal differences in carrying out activities, as each person shows
individual characteristics in their behaviour.
Simultaneous activities: activities performed by the user can be interwoven, as
people tend to multitask when one activity does not fully engage their
attention.
•
Incomplete execution of activities: when the user changes his or her plans, an
activity can remain incomplete.
•
Temporal ordering of the activity: the sequence of steps which is executed for
fulfilling an activity must be modelled. Furthermore, to recognize activities,
the preceding activities can give a strong indicator of what might be done next
by the user.
•
Mid- and long term trends: the distinction between variations of the execution
of an activity and significant deviations in a considered time frame has to be
recognised. A strong relation between the activity and the physical and
psychical constitution of the monitored person has to be taken into account. In
addition, not only a retrospective analysis, but also forecasting future
behaviour has to be realised.
•
Currently, for “ambient” sensors mounted in the environment, the largest amount
of activity recognition is carried out using cameras and computer vision techniques. For
example, Thiago Teixera et al. (Teixeira, 2006) describe a behaviour interpretation
framework that recognizes unsafe and out-of-the-ordinary human behaviour. It uses a
camera network covering the apartment and two types of patterns are observed. The
first are well- defined activities and rules that raise exceptions. The second type is
based on longer term statistical properties of behaviour. These are meant to recognize
shifts in behaviour patterns over a period of time. The network recognizes a set of
behaviour patterns and rules by reasoning using areas and locations.
5.2.5.
Reasoning approaches
For reasoning systems and choosing appropriate algorithms for activity recognition and
situation detection, the following requirements must be considered:
Handling imperfect information.
Information in the real world is subject to various
causes of imperfection like missing data, credibility of information sources and error in
measurement. Any system designed for reliable diagnosis and recognition in such an
environment has to take these aspects into account when combining information.
Handling temporal information.
Behaviour (both human and regarding technical
systems) is characterized by changes of (system) states over time. Formalisms for
modelling behaviour - as a first step towards recognizing it - therefore have to be
capable of expressing these temporal aspects.
Expressiveness of models.
The reasoning algorithm must work on underlying models
which are adequately expressive to describe complex situations and complex
interrelations between parameters in the description of human capabilities.
Scalability of models.
The reasoning algorithms must work in a scalable way in
respect of the model complexity and the number of significant parameters in the model.
