Target Identification and Distributed Cooperative Control of Sensor Networks

With the advances in communication and embedded systems, the monitoring and/or controlling of physical phenomena that span over wide spatial area have been attempted with deployment of a network of inexpensive and miniature sensors. In this paper, we focus on the automated sensor management for target identification at the application layer. The sensor management is formulated using graph grammar that reactively control the states of the sensors based on their proximity to the target and the states of their neighboring sensors. Target identification, on the other hand, concerns the estimation of the target's kinematics and attributes. The current practice is often formulated as finding the conditional probability of the target type on features derived from the sensor measurements with statistical pattern recognition. However, due to lack of training data, we demonstrate that the use of semantic latent indexing and stochastic approximation techniques, borrowed from the computer science community, is a more powerful method for sensor management and target identification.