Position Estimation Algorithms for Head-Mounted UWB Tags: A Comparative Analysis for Embedded Systems

Wearable Ultra-wideband (UWB) positioning systems are increasingly popular for indoor localization due to their resistance to interference and multipath signals. Those features enable improved accuracy over Bluetooth and WiFi on the other side UWB is still affected by noise and Non-Line-of-Sight (NLOS) conditions. Addressing these challenges, particularly in headmounted devices like smart glasses, requires the analysis of crucial factors such as NLOS interference from the head itself and optimizing algorithm choice for real-time performance. This paper quantifies the ranging errors in head-mounted UWB systems, considering the head as an NLOS element, and it evaluates robust on-device localization algorithms to identify the best options for energy-efficient, real-time onboard execution. According to our evaluation, the mean filter demonstrates superior accuracy (MAEs of 6.0-11.0 cm under moderate NLOS, and 10.3-16.1 cm with four NLOS anchors) and maintained robust performance across varying NLOS conditions. In contrast, extended Kalman Filter implementations, particularly those using IMU Inertial Measurement Unit data, exhibited higher error variability (MAEs of 13.8-18.2 cm under minimal NLOS to 19.6-22.9 cm in high NLOS scenarios) and longer processing times, requiring 1.94 ms per update, over 2.5 times longer than the mean filter's 0.76 ms per update. These findings underscore that simpler filtering approaches can balance accuracy, robustness, and computational efficiency in head-mounted UWB applications.