The overall task is quite daunting, and the opportunities to
develop proactive approaches to help with workflow management tasks are
However, RTLS deployments are still used in a relatively
basic way, as noted above, with little work focusing on how to leverage massive
indoor location traces.
To this end, this paper provides a focused study of workflow
modeling via integrated analysis of indoor location traces, evaluated on real
data from hospital environments.
Such workflow models serve as fundamental building blocks in
a wide range of workflow management problems.
In methods were proposed to discover periodic patterns from
spatio-temporal data, where a periodic pattern is defined as a regular activity
which periodically happens at certain locations.
Also proposed methods to discover sequential patterns from
imprecise trajectories of moving objects.
However, these methods were not developed for indoor spaces,
were not designed for the purpose of workflow modeling and, more importantly,
the mined frequent patterns cannot provide a parsimonious description of
healthcare activities in hospitals, o support the applications we have
Operation and Management
The learned workflow model is valuable, since a range of
practical problems can benefit from the modeling results.
Indeed, we have implemented a management information system
to exploit the discovered knowledge for healthcare operation and management.
Multiple locations may be used either concurrently or
interchangeably and, therefore, should be grouped together.
However, this grouping will depend on the workflow/procedure,
and may change over time.
We model the functional significance of a location in the
context of a workflow as a hidden state.
Therefore, a state is a probability distribution over rooms.
However, in order to obtain interpretable results, we need to regularize it.
Workflow modeling is central to many building operations and
management tasks, systematically constructing and estimating those models based
on massive indoor location traces is a nontrivial endeavor.
Next we identify specific challenges in workflow modeling
from indoor location traces, and we outline the ingredients of our proposed
solution, which revolve around representation of position location and of
mobility transitions at three different levels: micro, meso, and macro.
At the “micro” level, we have the raw data, which consist of
three-dimensional coordinates and geometric (Euclidean) distance between them.
Based on these, we have to construct or infer appropriate
representations for workflow modeling.