Research on Human Motion Recognition Based on Lower Limb Electromyography (EMG) Signals

Abstract

To explore the feasibility of identifying human movement by electromyography (EMG) signals of lower limb, we selected uphill, downhill, walking on flat ground and squatting as the movements and analyzed the joint angle at the initial stage of support phase. The length changes of lower limb muscles during gait were studied by OpenSim software. The lateral femoral, tensor fascia lata, biceps femoris, gastrocnemius and tibialis anterior muscle were selected as test muscles. The EMG signals in the process of exercise were collected. After filtered and extracted by the time domain (iEMG, RMS, and VAR) and the frequency domain (MF and MPF) analysis method, the EMG signals are input into BP neural network to identify movements. The results show that the periodic change of EMG signal is obvious in the process of exercise, the EMG signals of each muscle in the lower limbs are different in different movements. And the recognition rate of EMG signal for five kinds of exercise is up to 95.14% (normal walking), 91.29% (uphill), 90.99% (downhill), 91.53% (squat), and 98.82% (side squat), the average rate reach to 93.55%. The results provide a possibility of trend prediction of human motion intention.