Semi-Unsupervised Mitigation of Human Body Shadowing for Indoor UWB pedestrian tracking

in Ultra Wideband (UWB), large ranging errors occur under Non-Line-of-Sight (NLoS) conditions, which significantly degrades positioning accuracy. Human body shadowing (HBS) is a specific case of NLoS, which is a prominent error source for on-body UWB positioning. This work presents a tracking algorithm based on a Particle Filter (PF), designed to mitigate HBS-induced positioning errors by using an orientation-adaptive measurement model, consisting of a bank of Gaussian Mixture Models. The relative orientation is derived from Inertial Measurement Unit (IMU) data, and predicted positions from the tracking algorithm itself. We propose a second tracking algorithm in order to train the adaptive measurement model in a semi-unsupervised way, eliminating the need for accurate ground truth. The proposed algorithm outperforms a state of the art algorithm by an average of 11% (unsupervised) to 39% (supervised) in an experimental evaluation.