Robust Target Tracking in Sensor Networks with Measurement Outliers

Due to their enhanced reliability, surveillance coverage, and improved estimation accuracy, sensor networks have attracted significant interest in target tracking. In real applications, however, sensor measurements may contain outliers (i.e. measurements that differ significantly from the true values) because of sensor fault or impulsive disturbance. These outliers can result in reduced tracking precision or even tracking failure. To address this issue, we introduce centralized and decentralized robust tracking schemes that automatically identify and remove outliers in the sensor measurements. These schemes are based on an outlier detect-and-reject approach. At each sensor node, a new outlier-detection measurement model is developed by integrating its original measurement model with a binary outlier-indicator variable. Variational Bayesian inference is utilized to estimate both the states of the tracked target and outlier indicator variables.