Recognition of Daily Activities in Adults With Wearable Inertial Sensors: Deep Learning Methods Study.

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Human Movement Analysis

Background:

• Activities of Daily Living (ADL) are crucial for independence and quality of life.
• Traditional ADL assessment lacks precision and objectivity.
• Impaired ADL performance negatively impacts autonomy and well-being.

Purpose of the Study:

• To objectively evaluate human performance in ADL using wearable inertial sensors and AI.
• To overcome limitations of traditional subjective ADL assessment methods.
• To provide a tool for early dysfunction detection and personalized rehabilitation.

Main Methods:

• Developed wearable inertial sensors with accelerometers and gyroscopes.
• Created a database of 52,600 shoulder and back movement records.
• Trained 4 deep learning models for ADL recognition.

Main Results:

• Deep learning models achieved 95-97% accuracy, precision, recall, and F1-score.
• Models demonstrated a good balance between precision and recall.
• Bidirectional models showed slightly superior results and faster convergence.

Conclusions:

• Deep learning models effectively identify and classify daily activities noninvasively.
• Minimal sensorization ensures user comfort, adherence, and data reliability.
• This technology offers a significant advancement for clinical evaluation and rehabilitation of movement limitations.