Academic
Publications
Sensor scheduling for energy-efficient tracking in cluttered environments

Sensor scheduling for energy-efficient tracking in cluttered environments,10.1109/ITA.2011.5743561,George Atia,Venugopal Veeravalli

Sensor scheduling for energy-efficient tracking in cluttered environments  
BibTex | RIS | RefWorks Download
In this paper we study the problem of tracking an object moving randomly through a network of wireless sensors in the presence of clutter. Our objective is to devise strategies for scheduling the sensors to optimize the tradeoff between tracking performance and energy consumption. The presence of random interference introduces uncertainty into the origin of the measurements. Data association techniques are thus required to associate each measurement with the target or discard it as arising from clutter (False alarms). We cast the scheduling problem as a Partially Observable Markov Decision Process (POMDP), where the control actions correspond to the set of sensors to activate at each time step. Exact solutions are generally intractable even for the simplest models due to the dimensionality of the information and action spaces. Hence, we develop an approximate sensor scheduler that optimizes a point-based value function over a set of reachable beliefs. Point- based updates are driven by a non-linear filter that combines the validated measurements through proper association probabilities. Our approach efficiently combines Probabilistic Data Association techniques for belief update with Point-Based Value Iteration for designing scheduling policies. The generated scheduling policies, albeit suboptimal, provide good energy-tracking tradeoffs. I. INTRODUCTION Smart sensor management holds the potential to optimize the usage of sensor networks which typically operate on lim- ited resources. A significant body of research work considers tasking sensors in dynamically evolving environments for a wide range of applications including tracking, classification, monitoring and surveillance applications (1), (7), (9), (11), (13), (20). This paper focuses on sensor scheduling for track- ing in cluttered environments. Our goal is to design a central controller to schedule the sensors to track an object of interest in the presence of false alarms (clutter). A filtering component (linear or non-linear) is part of any sensor management algorithm, not only for the obvious purpose of state estimation, but also to provide the statistics necessary for the controller to select appropriate sensing actions and modes. Non-linear filtering for tracking in cluttered environments is particularly hard as it requires considering a large number of events due to the so-called data association problem, and is hence computationally intensive. The presence of random interference from nearby objects, false alarms, electromagnetic interference etc. generally leads to ambiguity in the origin of the sensor measurements and hence it is crucial to associate the measurements with their corresponding tracks. One simple and intuitive candidate solution for the association problem is to choose the signal with the highest intensity, among a set of validated measurements, for track update and discard the others. This is known as Strongest Neighbor Filter SNF (5). The Nearest Neighbor Filter NNF is another solution that uses the measurement closest to the predicted measurement obtained through a prediction step of the track estimation filter (2). However, these algorithms start to fail when the false alarm rate, or clutter density, increases. Alternatively, probabilistic data association (PDA) for a single target in clutter is another approach which uses all the validated measurements and does not discard any of them (3). A proper weight, reflecting the association probability, is assigned to each measurement and the weighted average of the validated innovations is used for the update. While most of the existing literature on target tracking in clutter has focused on the estimation aspect of the tracking problem using one or two sensors, the primary focus of this paper is on the design of efficient control policies organizing the activity of a larger network of sensors in the presence of false alarms. We cast the scheduling problem as a Partially Observable Markov Decision Process (POMDP), and devise strategies whereby the sensors are activated to optimize the fundamental tradeoff between energy expenditure and tracking performance in the presence of spurious measurements from clutter.
Cumulative Annual
View Publication
The following links allow you to view full publications. These links are maintained by other sources not affiliated with Microsoft Academic Search.