Feature Fusion of a Deep-Learning Algorithm into Wearable Sensor Devices for Human Activity Recognition

Area of Science:

• Biomedical Engineering
• Computer Science
• Machine Learning

Background:

• Accurate human activity recognition (HAR) is crucial for health monitoring and rehabilitation.
• Existing HAR systems often lack accuracy or require complex setups.
• Wearable devices offer a promising solution for unobtrusive activity monitoring.

Purpose of the Study:

• To develop and validate a wearable device for recognizing six daily activities using a deep learning algorithm.
• To assess the accuracy and reliability of the proposed HAR system.

Main Methods:

• A wearable device with a single-board computer and six-axis sensors was developed.
• A deep learning algorithm utilizing parallel convolutional neural networks (CNNs) with feature fusion was employed.
• The system was trained and validated using the UCI HAR dataset and self-recorded data from 21 participants.

Main Results:

• The system achieved high accuracy in recognizing six activities of daily living.
• Accuracy rates were 97.49% (UCI dataset) and 96.27% (self-recorded data).
• Tenfold cross-validation yielded accuracies of 99.56% and 97.46%, respectively.

Conclusions:

• The proposed CNN architecture demonstrates high performance for HAR.
• The wearable device is effective for activity recognition in diverse datasets.
• This technology has potential applications in rehabilitation assessment for individuals unable to perform strenuous exercise.