Enhanced Indoor Pedestrian Tracking Using UWB/PDR Fusion and NLOS Error Mitigation

Pedestrian Dead Reckoning (PDR) utilizing Inertial Navigation System (INS) technology often encounters substantial cumulative errors and random sensor-induced noise, while Ultra-Wideband (UWB) positioning can degrade severely under Non-Line-of-Sight (NLOS) conditions due to environmental factors. To overcome the above challenges, this research proposes an enhanced PDR/UWB fusion framework that significantly improves positioning accuracy. First, a novel NLOS ranging error mitigation approach is developed by integrating an Extreme Learning Machine (ELM) in conjunction with Gaussian Process Regression (GPR), with the initial parameters of the ELM optimized via the Nutcracker Optimizer. Channel Impulse Response (CIR) data collected from the UWB module serve as features, which are first processed by the ELM, and subsequently refined by the GPR, resulting in more accurate distance estimates under challenging NLOS conditions. Second, to accommodate scenarios where UWB signals are intermittently unavailable, a fusion strategy is proposed that employs a Multilayer Perceptron (MLP) neural network to predict pseudo-UWB observations from PDR data. Specifically, PDR-derived acceleration, angular velocity, and current position are fed into the MLP. When UWB measurements are absent, the MLP-predicted observations are integrated with PDR outputs using an Extended Kalman Filter (EKF), ensuring continuous and reliable localization. Experimental results demonstrate that the proposed integrated system substantially outperforms standalone UWB or EKF-based methods, achieving marked improvements in positioning accuracy and underscoring its effectiveness in delivering robust and precise indoor localization.