Abstract
Sensornets have been used for ecological monitoring the past decade, yet
the main driving force behind these deployments are still computer scien-
tists. The denser sampling and added modalities oered by sensornets could
drive these elds in new directions, but not until the domain scientists be-
come familiar with sensornets and use them as any other instrument in their
toolbox.
We explore three dierent directions in which sensornets can become easier
to deploy, collect data of higher quality, and oer more
exibility, and we postulate that sensornets should be instruments for domain scientists.
As a tool to ease designing and deploying sensornets, we developed a method-
ology to characterize mote performance and predict the resource consumption
for applications on dierent platforms, without actually having to execute
them. This enables easy comparison of dierent platforms.
In order to reduce the amount of faulty and missing measurements, we developed a mote-based anomaly detection framework lightweight enough to run alongside an actual data acquisition application. This allows faulty measurements to be detected immediately and not after the experiment has been concluded. To increase the exibility of sensornets and reduce the complexity for the domain scientist, we developed an AI-based controller to act as a proxy between the scientist and sensornet. This controller is driven by the scientist's requirements to the collected data, and uses adaptive sampling in order to reach these goals.
the main driving force behind these deployments are still computer scien-
tists. The denser sampling and added modalities oered by sensornets could
drive these elds in new directions, but not until the domain scientists be-
come familiar with sensornets and use them as any other instrument in their
toolbox.
We explore three dierent directions in which sensornets can become easier
to deploy, collect data of higher quality, and oer more
exibility, and we postulate that sensornets should be instruments for domain scientists.
As a tool to ease designing and deploying sensornets, we developed a method-
ology to characterize mote performance and predict the resource consumption
for applications on dierent platforms, without actually having to execute
them. This enables easy comparison of dierent platforms.
In order to reduce the amount of faulty and missing measurements, we developed a mote-based anomaly detection framework lightweight enough to run alongside an actual data acquisition application. This allows faulty measurements to be detected immediately and not after the experiment has been concluded. To increase the exibility of sensornets and reduce the complexity for the domain scientist, we developed an AI-based controller to act as a proxy between the scientist and sensornet. This controller is driven by the scientist's requirements to the collected data, and uses adaptive sampling in order to reach these goals.
| Original language | Danish |
|---|
| Place of Publication | København |
|---|---|
| Number of pages | 108 |
| Publication status | Published - 2009 |