Adaptive time series forecasting to restrain outliers for target tracking in wireless sensor networks

In this paper, we study the problem of outliers detection for target tracking in wireless sensor networks. Outliers are common in measurements because of the clutter environment, which bring significant errors to the estimate of target state and even result in filter divergence. In order to overcome this problem, this paper presents an adaptive time series forecasting method for restraining outliers. We first build an autoregressive model on each node to predict the next measurement, and then exploit Kalman filter to update the model adaptively, thus the outliers can be detected in accord with the deviation between the prediction by the model and the real measurement. The presented method is independent on the tracking algorithm and unaffected by the tracking accuracy. The simulation results show good performance in terms of effectiveness, robustness and tracking accuracy.