Database Reference
In-Depth Information
hour-long file representing the accelerometer deployed in an hour long moun-
tainous time trial (wickm), the second has values from an 18-hour long ultra
endurance race. As shown in the table, the query time for detecting cadence is
significantly reduced for the contextually enriched data. Of the 17,798 entries in
wickm.xml , 2,111 matched the criteria of a strong cadence. In the larger raim.xml
file, 7,631 of 65,536 entries correspond to a strong cadence. Following contextual
enrichment, we can detect vector magnitude of type = Low. Due to the incresed
number of results matching the criteria, the query time is longer than the query
for strong cadence.
The GPS based query is also performed both before and after contextual en-
richment. The time taken to evaluate is similar in both cases due to the reletively
small size of the input file, wickgps.xml . The main benefit of including GPS based
ranges as event definitions is that it allows the end user to specify the important
segments of the session which is applied directly to the data and made simple to
query. GPS coordinates are bulky and having to pass them as part of a complex
query to detect relevant segments of a session increases the potential for error.
As GPS coordinates differ for every environment, it is necessary for the end user
to have access to defining these boundaries eciently.
Table 3. Sample Enrichment Times
Event Filename Sensor Time
1 Cadence Classification wickm.xml GT3X Accelerometer 1,195ms
2 Cadence Classification raim.xml GT3X Accelerometer 12,245ms
3 Vector Magnitude Classification wickm.xml GT3X Accelerometer 1,074ms
4 Vector Magnitude Classification raim.xml
GT3X Accelerometer 11,309ms
5 Terrain Classification
wickgps.xml Garmin GPS
114ms
The time taken to contextually enrich the rules into the sensor files is dis-
played in Table 3, where times for vector magnitude classification and cadence
classification are proportional to the input filesize. While times can require up
to 12,245ms for the 30MB file, the process needs only to be performed once.
In summary, the experiments demonstrate that enablement and enrichment,
with their XML and semantic overheads, can be queried using high level query
languages without signifciant overhead. The main evaluation comes from our
collaborators, the exercise physiologists, who provide the datasets, specify the
queries, and can now extract information independently, using events and an
XQuery interface.
5 Related Research
[6] describes the approach to building OntoSensor, a prototype sensor knowledge
repository compatible with evolving Sensor Web infrastructure. OntoSensor in-
cludes definitions of concepts and properties adopted in part from SensorML, the
Web Ontology Language (OWL)[11] and extensions to IEEE Suggested Upper
 
Search WWH ::




Custom Search