Robust, Distributed Target Tracking Using Sensor Network

Kartavya Neema, Purdue University

Abstract

Distributed target tracking using sensor networks is crucial for supporting a variety of applications such as battlefield monitoring, weather monitoring, and air traffic management. This dissertation presents a problem formulation and solution approach for distributed target tracking, comprising of sensor fusion and sensor target allocation problems, in the presence of faults in the sensor measurements. There are times when an architecture with central node is preferred but other times when distributed is necessary, we seek a distributed case that can approach the attractive features of centralized case. Therefore, we propose that the underlying two-fold goals of the distributed target tracking problem is to: (1) reach a consensus in the allocation decisions across the sensor network, and (2) achieve a consensus in the state estimates across all the sensors in the network. These goals ensure that each sensor node has the same information across the sensor network, and any node can behave as a central node. In the process of achieving our goals, we develop two new algorithms, one for distributed sensor-target allocation and another for distributed sensor fusion. The Dual Phase Consensus Algorithm (DPCA) for distributed sensor target allocation is a real time algorithm that works in two phases. The first phase of DPCA is similar to distributed sequential greedy search that combines the benefits of greedy and consensus algorithms to reach a feasible solution. The second phase iteratively improves the allocation eventually leading toward a global optimum. DPCA converges to a feasible solution at the order of number of sensors, and thus can be useful for implementation in real time systems. For distributed sensor fusion, we extend the state-of-art distributed Kalman filtering technique called Generalized Kalman Consensus Filter (GKCF), and make it robust against faults present in the sensor measurements. We particularly focus on two types of faults: (1) outliers in the sensor measurements, and (2) unknown change in covariance of the sensor measurements. A combination of analytical and hardware redundancy techniques are used to detect and mitigate the effects of faults. Analytical redundancy is based on generating residues (difference between measured and estimated output) that are sensitive to the presence of particular type of faults, whereas hardware redundancy detects faults by the comparison of the same measurement via two different hardwares (sensors for our case). Using these two techniques, a new algorithm called Robust Generalized Kalman Filter (RGKCF) is developed that achieves a consensus in the state estimates across the sensor network in the presence of faults. Finally, we propose a combination of two algorithms, DPCA and RGKCF, to solve the problem of robust, distributed target tracking. RGKCF helps the sensor to move toward a consensus in the state estimates in the presence of faults in the sensor measurements, and DPCA achieves a consensus in the allocation decisions. Using simulation results, we demonstrate the effectiveness of the proposed approach to satisfy our two-fold goals, and also compare it with a centralized approach. We further show that the distributed approach is sensitive to network topology and suggest network diameter as a measure to provide guideline to architect a sensor network for diverse distributed target tracking applications.

Degree

Ph.D.

Advisors

DeLaurentis, Purdue University.

Subject Area

Aerospace engineering

COinS