Sensor deployment in detection networks-a control theoretic approach

Abstract

For any automated surveillance operation to be successful, it is critical to have sensing resources strategically positioned to observe, interpret, react and maybe even predict events.In many practical scenarios, it is also expected that different zones within a surveillance area may have different probability of event detection (or false alarm) requirements. The operational objective in such surveillance systems is to optimize resources (number of sensors and the associated cost) and their deployment while guaranteeing a certain assured level of detection/false alarm performance. In this dissertation, we study two major challenges related to sensor deployment in distributed sensor networks (DSNs) for detection applications. The first problem we study is the sensor deployment problem in which we ask the following question: Given a finite number of sensors (with a known sensing profile), how can we deploy these sensors such that we best meet the detection and false alarm requirements in a DSN employing a specific information fusion rule? Even though sensor deployment has garnered significant interest in the past, a unified, analytical framework to model and study sensor deployment is lacking. Additionally, the algorithms proposed in literature are typically heuristic in nature and are limited to (1) simplistic DSN fusion architectures, and (2) DSNs with uniform detection/false alarm requirements. In this dissertation, we propose a novel treatment of the sensor deployment problem using concepts from optimal control theory. Specifically, the deployment problem is formulated as a linear quadratic regulator (LQR) problem which provides a rigorous and analytical framework to study the deployment problem. We develop new sensor deployment algorithms that are applicable to a wide range of DSN architectures employing different fusion rules such as (1) logical OR fusion; (2) value fusion; (3) majority decision fusion, and (4) optimal decision fusion. In all these cases, we demonstrate that our proposed control theoretic deployment approach is able to significantly outperform previously proposed algorithms. The second problem considered in this dissertation is the “self healing” problem in which we ask the following question: After the failure of a number of sensors, how can one reconfigure the DSN such that the performance degradation due to sensor loss is minimized? Prior efforts in tackling the self healing problem typically rely on assumptions that don’t accurately capture the behavior of practical sensors/networks and focus on minimizing performance degradation at a local area of the network instead of considering overall performance of the DSN. In this work, we propose two self healing strategies the first approach relies on adjusting decision thresholds at the fusion center. The second approach involves sensor redeployment based on our control theoretic deployment framework. Simulation results illustrate that the proposed algorithms are effective in alleviating the performance degradation due to sensor loss.