Gait Phase Detection Based on a Foot-Mounted Inertial Sensor Using Long Short-Term Memory Enhanced by Hidden Markov Model

Gait phase detection provides necessary temporal parameters for gait analysis. Accurate but low-cost detection of gait phase makes gait analysis reliable and easy to use. In this work, we apply hidden Markov model (HMM) to improve gait phase detection based on long and short-term memory (LSTM), and propose a hybrid LSTM/HMM model. Our proposed model detects four gait phases by taking signals from a foot mounted inertial sensor as input. Sensitivity and specificity are adopted to evaluate our proposed model on five different terrains including level ground, ramp ascent, ramp decent, up stair and down stair terrains. From the results, the sensitivity and specificity of the hybrid LSTM/HMM model significant outperformed than those of the LSTM-based model for gait phase detection, and reached 96.83% and 99.08%, respectively.