Challenges of Trajectory Estimation using Inertial Sensors in Smartwatches

This paper presents a comprehensive framework for estimating running and walking trajectories using inertial sensors in smartwatches, integrating and expanding upon our previous research. Due to the limitations of GNSS accuracy in smartwatches, especially in environments like urban canyons, we propose a smartwatch-based pedestrian dead reckoning (PDR) approach to address these challenges. Our framework combines key components: step detection, which dynamically selects the appropriate sensor signal (e.g., angular velocity or acceleration) based on the user’s activity; movement direction estimation through a kinematic model and principal component analysis (PCA) to correct arm swing misalignment; and stride length estimation using a BiLSTM/CNN-based deep learning model with a data augmentation strategy. Experimental results from walking and running exercises demonstrate that the proposed smartwatch-based method significantly improves trajectory estimation accuracy compared to GNSS, particularly in urban environments.