A Low-Cost End-to-End sEMG-Based Gait Sub-Phase Recognition System

Area of Science:

• Biomedical Engineering
• Human Movement Analysis
• Wearable Technology

Background:

• Surface electromyogram (sEMG) signals are crucial for detecting human movement intentions and are used as control inputs.
• Current sEMG-based gait recognition systems face challenges with manual labeling, bulky equipment, and poor portability.
• There is a need for cost-effective, portable, and accurate systems for gait sub-phase classification.

Purpose of the Study:

• To develop a low-cost, end-to-end system for gait sub-phase recognition using sEMG signals.
• To integrate wireless sEMG acquisition with a novel LSTM-MLP neural network classifier.
• To evaluate the system's performance under various walking conditions.

Main Methods:

• A wireless multi-channel sEMG acquisition device was developed to collect thigh muscle sEMG and plantar pressure signals.
• A novel neural network classifier combining Long Short-Term Memory (LSTM) and Multilayer Perceptron (MLP) was designed.
• The system was evaluated on subjects walking under five conditions: varied speeds, inclines, and carrying loads.

Main Results:

• The proposed system achieved high average classification accuracies: 94.10% (flat, 5 km/h), 87.25% (flat, 3 km/h), 90.71% (20 kg backpack, 5 km/h), 94.02% (20 kg shoulder bag, 5 km/h), and 87.87% (15° slope, 5 km/h).
• These accuracies were significantly higher than existing gait recognition methods.
• The system demonstrated excellent real-time performance with an average inference time between 3.25 and 3.31 ms.

Conclusions:

• The developed system offers an effective and low-cost solution for sEMG-based gait sub-phase recognition.
• The combination of wireless sEMG acquisition and the LSTM-MLP classifier significantly enhances recognition accuracy and portability.
• This technology has the potential to advance wearable gait analysis and human-computer interfaces.