Heterogeneous Sensor Webs for Automated Target Recognition and Tracking in Urban Terrain (HSN)
S. Shankar Sastry, Michael Gastpar, Avideh Zakhor, Trevor Darrell and Jitendra Malik
Army Research Office W911NF-06-1-0076
Over the past five years, network embedded systems and sensor networks have emerged as an important new class of computation that embeds computing in the physical world. The continued miniaturization of processors and storage, combined with the emergence of low power communications such as CMOS radios and mass-produced sensors (using MEMS) has enabled complete highly integrated nodes in easily-deployed, self-calibrating, disposable sensor networks. Such networks gather multi-modal data about the environment, use both local and distributed inference algorithms to determine reliable interpretations at multiple levels of granularity, and communicate those interpretations in response to events or queries.
After numerous experiments and demonstrations, the following issues have been identified as major factors limiting scaling, robustness, and the level of usability:
- Sensor bandwidth;
- An inadequate theory of distributed signal processing;
- Lack of robustness;
- Performance metrics; and
- Lack of mobility of the sensor networks.
The above issues will be addressed in this research project and the high-level objectives of this research project are to develop:
- A new theory of distributed signal processing with random spatio-temporal sampling of complex scenes for recognition and tracking of objects;
- Robust design principles for sensor networks with both low- and high-bandwidth sensors, organized to be able to automatically recognize and track targets in complex urban environments;
- Metrics for the design and deployment of sensor networks. We will derive theoretical bounds on the performance of different kinds of sensor webs (based on the density of their deployment, their manner of deployment and the choice of sensing, networking and signal processing algorithms); and
- Incorporate mobility into sensor networks. We will develop algorithms which take into account mobility of nodes and the need to query sensor nodes using mobile assets such as UAVs.