|
|
||||
|
|
|
|||
| Overview |
As sensor network technology advances further, it will become cheaper and easier
to deploy more sensors in more diverse application scenarios. However, it is
still extremely difficult for scientists and engineers to design signal
processing algorithms to accurately extract events from the raw sensor data. We
propose a system called Sieve that uses unsupervised learning to
automatically identify recurring events in sensor streams, reducing the time
required for event extraction. We used this system to differentiate between
walking, running, jumping, sitting, standing, and falling based on accelerometer
data, and to recognize activities such as cooking and using the bathroom based
on data collected from home sensors. We compared Sieve to supervised learning
and found that both provide comparable success at event extraction, while Sieve
has better scaling properties.
| |||
| Publications |
Timothy Hnat and Kamin Whitehouse.Sieve: Simple Event Classification for Wireless Sensor Networks. Under submission. |
|||
|
|
|
|||
|
Kamin Whitehouse Computer Science Department The University of Virginia 217 Olsson Hall Charlottesville, Virginia 94720 |