A hybrid neural network approach to gait analysis for remote patient care

Computer Vision based Gait Analysis (GA) has evolved Remote Patient Monitoring (RPM), by estimating key gait parameters (GPs) from live videos. Estimation of GPs is evaluated through four different algorithms, including state-of-the-art hybrid LSTM-Transformer. Three accuracy parameters are being utilized i.e., correlation, mean absolute error (MAE) and mean absolute percentage error (MAPE). Hybrid LSTM-Transformer stands out with MAEs of 0.1397 m/s for speed, 0.090 steps/s for cadence, 5.532° for knee flexion, and 6.25 for GDI; MAPE values of 20.81% (speed), 11.00% (cadence), 3.50% (knee flexion), and 8.00% (GDI); and correlations of 0.791, 0.790, 0.851, and 0.753 for the GPs estimation.