|
Wireless sensor networks are being succesfully applied to
environmental control and monitoring. A dense deployment of low-cost
low-power sensors in the region of interest (Fig. 1) can provide
information, like temperature or atmospheric pressure,
at fine scales.
This information is transmitted from each sensor through the network and stored at a central node
in order to be analysed by the end users.
Fig. 1: Generic sketch of a wireless sensor network for
environmental monitoring
Adding vision capabilities
to the nodes increases the potential of the network quite significantly.
In particular, the detection of dynamic textures is specially interesting
in monitoring natural scenarios. A dynamic texture is a spatially-repetitive
time-varying visual pattern whose temporal variation presents certain stationarity.
An additional feature of a dynamic texture is its indeterminate spatial and temporal extent.
Smoke colums (Fig. 2), wave patterns, swaying trees or a flock of birds are some examples
of dynamic textures.
These examples point out that dynamic textures are very common in natural scenes.
Fig. 2: Smoke, an example of dynamic texture
The main effort in this project is focused on designing a vision chip
intended to segment and detect dynamic textures by means of bioinspired energy-efficient
focal-plane processing. The vision system-on-a-chip will be able also to realize complex
algorithms with a more conventional processor.
Finally, as a direct application of such a chip, we have developed a new approach for very
early detection of forest fires by means of visual smoke detection.
|
