Database Reference
In-Depth Information
1.2 Contribution
In simple terms, data can be queried if we develop a protocol to transfer it into
a relational database or encode the data in XML. However, through working
with both exercise physiologists and cyclists, we discovered that their informa-
tion needs could not be met (queries could not be expressed) with a process
of supplying structure and low level semantics to sensor data. Instead, a more
complex layer of contextual enrichment was required to prepare sensor data for
high level query languages. Furthermore, this contextual enrichment must be
specified by end users and not by computer scientists. In this paper, we present
a framework and methodology for automated processing of sensor data so that
it can be queried using a standard query language. While this method uses XML
to provide the structre for sensor data, it is the end user (domain expert) who
can add semantics to the data through the specification of data mining rules. By
working closely with end user scientists, we evalute our system by meeting the
information needs of the end user, allowing them to specify how data repositories
are enriched with context data, and by reducing the query execution times as a
result of the contextual enrichment process.
1.3 Structure
The structure of the paper is as follows:
2 introduces cycling, the domain in
which our system was deployed and provides an overview of the EventSense
system architecture, with the Context Profiles explored in detail in
§
3.
4 details
§
§
our experimental evaluation and results, and in
§
5 we present related research.
§
6 details our conclusion and our current work.
2 User Requirements and Operating Architecture
In this section, we present the user requirements in the form of a query set,
defined by the end users. Queries 1 to 5 in Table 1 can be expressed using
XQuery but the remaining queries are more complex, dicult to express and
may require long calculation times. We will then describe the architecture used
enable the exercise physiologists to extract the required information.
In general, the system must collect data from several independent sources, syn-
chronise the data, and structure the data in some manner. It must also provide
a facility for defining and applying event rules specific to a particular domain. In
Table 1, this includes the hill classification, and complex accelerometer based al-
gorithms for pedal cadence/vector/force and braking activity. The system needs
to work in a context driven environment where the user can specify if the data
comes from a training session, race, or the laboratory.
2.1 The EventSense Architecture
Figure 1 illustrates the architecture of our proposed solution to sensor data
management. The remainder of this section details the individual processors
