The Adaptive Tobit Kalman Filter: Tracking Position with Censored Measurements in the IoT (1909.05603v1)

Abstract: In the Internet of Things (IoT) paradigm, distributed sensors and actuators can observe and act on their environment, communicating wirelessly. In this context, filtering the observations and tracking the network and environment state over time is extremely important, and the Kalman Filter (KF) is one of the most common tools for this. Several of these applications deal with censored data, either because of sensor saturation or limited detection regions: when censoring happens, all measurements below a certain threshold are clipped to the threshold value. The recently proposed Tobit Kalman Filter (TKF) is an adjusted version of the KF that can deal with censored measurements. However, like the traditional KF, it needs full knowledge of the process and measurement noise covariances to work, which are not always available in practice. In this work, we relax this assumption and propose the Adaptive Tobit Kalman Filter (ATKF), which can dynamically estimate the process and measurement noise along with the hidden state of the system from the censored measurements. We apply our solution to navigation and positioning IoT scenarios, obtaining a negligible performance loss with regard to the TKF, even with no a priori knowledge of the noise statistics.