Fault tolerant target tracking in sensor networks

Abstract

In this paper, we present a Gaussian mixture model based approach to capture the spatial characteristics of any target signal in a sensor network, and further propose a temporally-adaptive variant of the approach for dynamic multiple target tracking under changing environments, with the presence of both significant background event noises and a large portion of outlying sensor readings. The target position is estimated by adopting the mean-shift optimization to discriminate the target signals from the background noises. Our mixture model based algorithm is capable of fusing multivariate real-valued sensor measurements and its probability nature shows fault tolerance and robustness in noisy sensing environments. This consideration is practical as in real world applications, sensor readings are multi-modal and may contain errors. The simulation study validates our design and the results indicate that our mixture model based algorithm is an effective and capable approach for the two most typical target signal models under consideration. Desirable quantitative target tracking results are also achieved through extensive evaluations under challenging background conditions.