Human gait recognition using LSTM
DescriptionA novel wearable solution using soft robotic sensors (SRS) introduced in previous parts of this series has been investigated in this study to model foot-ankle kinematics during gait cycles. Employing eight SRS attached to a pair of compression socks, the capacitance of SRS related to four basic movements including plantar flexion (PF), dorsiflexion (DF), inversion (INV), and eversion (EVR) for the left and right foot were quantified during the gait movement of 20 participants. Three-dimensional (3D) motion capture data was also collected for analyzing gait movement in order to evaluate the power of SRS in modeling foot-ankle kinematics. Experiments were designed to be placed on two different walking surfaces including a flat surface and a cross-sloped surface with the intention of simulating the dynamics of real-life gait movements. Long Short-Term Memory (LSTM) network was employed to quantify the relationship between SRS and the 3D motion capture system. Models performance was evaluated based on Root Mean Squared Error (RMSE) of prediction of joint angle of the foot in the sagittal and frontal plane, collected from the motion capture system. LSTM achieved very low RMSE error between 2.48 and 5.74 degrees for different modeling designs which reveals the high performance of SRS in capturing foot-ankle kinematics during the human gait cycle.