Embedded sensor networks

A wireless sensor network (WSN) is made of numerous small independent sensor nodes, which consist of a battery, radio, sensors, and a minimal amount of onboard computing power and has wireless communication capabilities among sensor nodes. Application of sensor nodes is quite common in the field of Agricultural monitoring (for climate sensitive crops), structural monitoring etc. However object detection, classification and tracking using WSN is more challenging and involves sensing and recognizing an object that enters the network area by collecting signals (acoustic, magnetic etc.) and processing them not only in real time but also involving restricted power consumption constraints. The real time applications include vehicle tracking, inventory tracking inside factories, military tracking and identification of hostile intrusion vehicles, and automatic tollgate collection based on two wheelers or four wheelers.

Sensor nodes are distributed in an optimized pattern and by taking in to account the limited power sources of sensor nodes, the tracking of object or target with random movement patterns is done by an energy efficient tracking algorithm, called selective activation combined with prediction. Audio signals are used to detect and classify the objects among the known classes. Specific parameters called feature vectors are calculated from the sampled audio signals for various objects and are used to train the classifier for classifying the object to its class. The selection of appropriate classifier and feature vectors proved to be very significant for achieving high accuracy percentage in classification. Thus a highly customized version of classifier and tracking algorithm specifically suitable for WSN is achieved.

The proposed classification and tracking algorithm is implemented in matlab considering vehicles (Light and Heavy Vehicles) as targets and audio signal as signal source. The classification of the target is done along predictive tracking and the energy consumed per unit time in the sensor network is calculated. The overall response of the classifier for varied signals is also tested and accuracy percentages were determined. Thus the project forms a complete solution for implementation of intrusion detection, identification and tracking of targets using wireless sensor network technology.