Lower Limb Gait Estimation Using Foot Motion and Neural Network

Gait analysis plays a crucial role in detecting gait abnormalities and providing technical assistance in clinical rehabilitation. However, traditional gait analysis methods, which rely on gold standard gait acquisition systems, are limited to indoor activities. On the other hand, wearable sensors like Inertia Measurement Units (IMUs) offer convenience but are prone to noise disturbance. To enable long-term gait analysis, recent trends have shifted towards developing Machine Learning (ML)-based gait analysis approaches. A significant portion of current research focuses on predicting joint kinematics, gait classification, and gait acceleration using ML techniques. This paper specifically explores the use of Long Short-Term Memory (LSTM) network for gait estimation. The approach taken in this paper involves utilizing Ground Reaction Forces (GRF) to estimate joint moments at the hip, knee, and ankle. The LSTM model achieved a correlation score of 0.86, indicating its potential for accurately estimating joint moments. Furthermore, employing a blend of CNN, LSTM, and GRU (Gated Recurrent Unit) has the potential to enhance estimation accuracy.