This is a repository copy of Spatio-temporal Gaussian process models for extended and group object tracking with irregular shapes.

Received signal strength based device-free localization applications utilize a model that relates the measurements to position of the wireless sensors and person, and the underlying inverse problem is solved either using an imaging method or a nonlinear Bayesian filter. In this paper, it is shown that Bayesian filters nearly reach the posterior Cramér-Rao bound and they are superior with respect to imaging approaches in terms of localization accuracy because the measurements are directly related to position of the person. However, Bayesian filters are known to suffer from divergence issues and in this paper, the problem is addressed by introducing a novel Bayesian filter. The developed filter augments the measurement model of a Bayesian filter with position estimates from an imaging approach. This bounds the filter's measurement residuals by the position errors of the imaging approach and as an outcome, the developed filter has robustness of an imaging method and tracking accuracy of a Bayesian filter. The filter is demonstrated to achieve a localization error of 0.11 m in a 75 m 2 open indoor deployment and an error of 0.29 m in a 82 m 2 apartment experiment, decreasing the localization error by 30 − 48 % with respect to a state-of-the-art imaging method.

No abstract

This paper discusses a novel approach to vehicle property sensing based on traffic induced road surface vibrations and investigates the feasibility of this approach. Road surface vibrations from real-life experiments are acquired using 3-axis accelerometers and the data is analyzed. Based on the assessment of the data, a first coarse scheme for axle detection of passing vehicles is developed. The scheme is then evaluated using measurement data from a highway with moderate traffic intensity but diverse traffic. It is found that the proposed approach is feasible and the estimation scheme yields promising results. Furthermore, delimitations, encountered problems and identified research challenges are discussed and future research directions are given.

Received signal strength (RSS)-based device-free localization applications utilize the communication between wireless devices for locating people within the monitored area. The technology is based on the fact that humans cause changes in properties of the wireless channel which is observed in the RSS, enabling localization of people without requiring them to carry any sensor, tag or device. Typically this inverse problem is solved using an empirical model that relates the RSS to location of the sensors and person, and utilizing either an imaging method or a particle filter (PF) for positioning. In this paper, we present an extended Kalman filtering (EKF) solution that incorporates some of the beneficial properties of the PF but has a lower computational overhead. In order to make the EKF work, we also need to reconsider how the measurements are sampled and processed, and a new processing scheme is proposed. The developments are validated using simulations and experimental data, and the results imply: i) the non-linear filters outperform a popular imaging method; ii) the robustness of the EKF and PF is improved using the proposed processing scheme; and iii) the EKF achieves similar performance as the PF as long as the new processing scheme is used.