Information Technology Reference
In-Depth Information
5.2.4.
Activity / behaviour recognition
The reasoning tasks for the recognition of situation and behavior can be differentiated
in five different levels of semantic abstraction:
Basic physical activity.
The basic physical activity of the user in terms of motion is
detected. This includes basic motion information from location tracking, movement of
the hand, body motion or general activity.
General activity.
The characteristics of physical activity are evaluated for describing
general activity towards the modes of locomotion (e.g. sitting, standing walking or
lying), and the user's location within the environment, on room level, on the level of
functional areas in the environment or in spatial relation to objects in the environment.
Specific actions.
Specific actions performed by the user are inferred from punctual
observations, describing interactions of the user with the environment (e.g. opening the
front door and leaving the house) or specific events in motion (e.g. falling, stand-up-
and-go).
Complex processes.
Most activities of daily living follow a planned sequence of action
steps, which are executed in order to fulfil a certain goal. To monitor these complex
processes, multiple actions and events must be correlated in respect to their temporal
ordering. Further information about these activities can be gathered by regarding their
effects on the environment (e.g. when the user is showering, the humidity level in the
bathroom will increase). The complexity for modelling and recognition of these
processes increases as the fact that the same person often performs the same activity
differently, and also different people perform the same activity differently has to be
taken into account.
Behaviour analysis.
To detect relevant situations medically, changes in the user's
behaviour and vital data must be inferred and evaluated to the extent that they are
affected by changes in the physical health status of the user. From analysing trends in
the user's activities and vital data, deteriorations in the physical health status can be
derived. Another impact factor is the circadian rhythm in the performance of daily
activities. Detected activities are evaluated against “normal” behaviour that a user
shows during the day, in order to infer changes in the habitual schedule of the user.
The complexity of the reasoning task increases with each level of abstraction, on
the one hand by the complexity for the modelling of activities and situations, and on
the other hand by the amount of heterogeneous information that must be fused in a
situational way. Challenges for activity recognition which must be met by the
reasoning system are:
•
Variations in the execution of activities: people tend to perform the same
activity in different ways. For example, the preparation of meals can cover the
preparation of a quick snack or extended cooking. These intrapersonal
differences in carrying out activities affect the type and order in the steps that
are performed, as well as the length of time in which the user spends on this
activity. Besides this, the reasoning system must also be able to adapt to
