ULTRA: UWB-based Localization and TRAcking through A-DBSCAN and Bayesian Algorithm

This paper aims to provide a sub-meter localization and tracking capability for vehicle key fob detection in challenging outdoor environments. We take an experimentation-oriented approach and design a UWB-based real-time localization and tracking system called ULTRA. We introduce two novel algorithmic components to improve the accuracy and reliability: (i) We first generate a set of potential locations based on their geometric features, then propose an algorithm based on A-DBSCAN (adaptive density-based spatial clustering of applications with noise) to dynamically form clusters, estimate sensor trustworthiness and the respective probabilities. (ii) We design a Bayesian algorithm without requiring extra motion sensors, which seamlessly balances the real-time measurements with history-based motion tracking. Thanks to these novel algorithms, ULTRA is able to better handle signal blockage, uncertain ranging noises caused by non-line-of-sight (NLOS) caused by nearby vehicles and parking structures. Using a combination of vehicle testbed systems and high-fidelity emulation platforms, we experimentally demonstrate that ULTRA has a localization accuracy of ≤30cm, and achieves a directional angle error of ≤ 5° under challenging outdoor environments, outperforming the state-of-the-art linear/nonlinear least squares algorithms, as well as other unsupervised-learning-based approaches (i.e., K-means clustering, SVM unsupervised outlier detection).