Exploration of Human Activity Recognition Using a Single Sensor for Stroke Survivors and Able-Bodied People
Long Meng1, Anjing Zhang2, Chen Chen1,3
1Department of Electronic Engineering, School of Information Science and Technology, Fudan University, Shanghai 200438, China.
Sensors (Basel, Switzerland)
|February 3, 2021
Summary
Accelerometers offer the best performance for human activity recognition (HAR) across different user groups, including stroke survivors. Optimal sensor placement, like on the extensor carpi ulnaris, is crucial for achieving high accuracy in HAR tasks.
Area of Science:
- Biomedical Engineering
- Rehabilitation Technology
- Human Activity Recognition
Background:
- Human Activity Recognition (HAR) commonly utilizes sensors like accelerometers, gyroscopes, and surface electromyography (sEMG).
- Limited research exists on optimal sensor selection and placement for effective HAR, particularly across diverse populations.
- Understanding sensor efficacy is vital for developing practical HAR solutions, especially for individuals with mobility impairments.
Purpose of the Study:
- To comparatively evaluate the performance of accelerometers, gyroscopes, and sEMG sensors for HAR.
- To investigate the impact of sensor placement on HAR accuracy across able-bodied individuals and stroke survivors.
- To propose a novel HAR approach for stroke survivors using models pre-trained on healthy subjects.
Main Methods:
- Comparative analysis of accelerometer, gyroscope, and sEMG sensor data for classifying four activities: walking, tooth brushing, face washing, and drinking.
- Utilized Support Vector Machine (SVM) classifier with Leave-One-Subject-Out Cross-Validation (LOSO-CV) for performance evaluation.
- Assessed sensor performance across three subject groups: able-bodied, stroke survivors, and a combined group.
Main Results:
- Accelerometers consistently demonstrated superior performance across all evaluated groups.
- The highest HAR accuracy for stroke survivors was 95.84 ± 1.75%, achieved with an accelerometer on the extensor carpi ulnaris.
- A novel approach for stroke survivor HAR, using a pre-trained model from healthy subjects, yielded a highest accuracy of 77.89 ± 4.81% with an accelerometer on the extensor carpi ulnaris.
Conclusions:
- Accelerometers are the most effective sensors for HAR, offering robust performance across different populations.
- Optimal sensor placement significantly influences HAR accuracy, with the extensor carpi ulnaris identified as a key position for stroke survivors.
- Transfer learning from healthy subjects to stroke survivors shows promise for developing practical HAR applications in rehabilitation.
