Subject-Independent Ground Reaction Force Estimation Using Smart Lacelock Data

Abstract

Ground reaction force (GRF) is a fundamental biomechanical parameter with critical applications in gait analysis, abnormality detection, and control of assistive devices. Traditional GRF measurement tools, such as force plates and instrumented treadmills, are accurate but impractical for daily use due to their cost, size, and laboratory constraints. This study presents a subject-independent method for estimating GRF using data from a novel wearable device, the Smart Lacelock, which integrates an inertial measurement unit (IMU) and a loadcell into a lightweight, shoelace-mounted sensor. A bidirectional long short-term memory (BiLSTM) neural network model was trained and evaluated using leave-one-out (L1O) cross-validation on data collected from 20 subjects. Forward feature selection identified vertical acceleration (acc_z) as the most significant feature, which was then used to train the BiLSTM model to estimate the stance phase of the GRF cycle. The model was evaluated using Pearson correlation, normalized root-mean-square error (nRMSE), and peak-based error measures. The model achieved mean Pearson correlations of 91% (left shoe) and 90% (right shoe), with corresponding nRMSE values of 0.24 and 0.27. Peak height and location differences averaged 12.87% and 5.13% for the left shoe, and 14.35% and 4.81% for the right shoe, respectively. These results demonstrate that the Smart Lacelock device, paired with the deep learning framework, can offer a practical, cost-effective, and accurate solution for GRF estimation in real-world environments.