Feature discovery and sensor discrimination in a network of distributed radar sensors for target tracking

Abstract

A spatially distributed network of radar sensors is being used for target tracking and for generating a single integrated aerial picture (SIAP). In such a network generally each sensor sends whatever target track/association information it has to every other sensor. This has the disadvantage of requiring more communication bandwidth and processing power. One of the ways to reduce the communication bandwidth and the processing power is to discover features that would improve the target detection/track accuracy and activate those sensors that would provide the missing information and, form clusters of sensors that have consistent information. We describe a minimax entropy based technique for feature discovery and within class entropy based technique for feature/sensor discrimination. After discovering the features, those sensors that can provide the discovered features are activated. The decision based on the sensor discrimination is used in cluster formation. The experimental details and simulation results that are provided here indicate that these metrics are efficient in discovering features and in discriminating sensors. The techniques described are dynamic in nature - as it acquires information it is making a decision on whether it is from a good sensor in terms of consistency. This has the advantage of discarding non-valid information dynamically and making progressive decision.