Distributed target tracking and boundary estimation in wireless sensor networks

We study the problem of target tracking and boundary detection of a substance diffusing from a mobile source (e.g., spread of a toxic gas from a moving vehicle) using a wireless sensor network. We present a predictionbased mobility adaptive tracking (P-MAT) algorithm to study the tradeoff between energy, accuracy of tracking, coverage, and boundary estimation. P-MAT minimizes overall energy consumption by incorporating both advance resource reservation and adaptive tracking. It uses adaptive Kalman filtering to predict the target’s future location and velocity. The predicted target location determines a set of sensors surrounding that location to be activated known as the active region. Sensors in the active region are responsible for target tracking and boundary detection. The predicted velocity determines the rate at which sensors sample the target. In summary, for tracking, P-MAT incorporates adaptivity in two forms: (1) the size of the active region and (2) modulation of the sampling rate. In this paper, we include dynamic boundary estimation. Boundary estimation in many situations can be performed efficiently using a subset of nodes within the vicinity of the phenomenon. This subset of nodes in our algorithm is the set of nodes in the active region. As the substance spreads, sensors in the active region determine if additional sensors outside of the active region are needed to enclose the boundary. We introduce two deployment models, the expected k-coverage model and the minimum k-coverage model. The goal of these k-coverage models is to increase the detection probability and accuracy of tracking within the target area at all times. Results from simulation experiments show that P-MAT can perform both tracking and boundary estimation with a desired accuracy while minimizing energy consumption.