Human Activity Prediction Based on Forecasted IMU Activity Signals by Sequence-to-Sequence Deep Neural Networks.

Area of Science:

• Computer Science
• Biomedical Engineering
• Machine Learning

Background:

• Human Activity Recognition (HAR) traditionally uses past sensor data.
• Predicting future human activities (HAP) is less explored but crucial for applications like fall detection.
• Existing HAR systems lack the predictive capabilities for proactive interventions.

Purpose of the Study:

• To develop and evaluate a novel Human Activity Prediction (HAP) system.
• To leverage forecasted Inertial Measurement Unit (IMU) data for predicting future human activities.
• To enhance proactive safety and health monitoring through accurate activity prediction.

Main Methods:

• A deep learning forecaster based on Sequence-to-Sequence architecture with attention and positional encoding was developed.
• A pre-trained deep learning Bi-LSTM classifier was employed to recognize future activities from forecasted IMU data.
• The system was tested using two tri-axial IMU sensors for five daily activities.

Main Results:

• The forecasted IMU signals demonstrated a high average correlation of 91.6% with actual measured signals.
• The HAP system achieved an impressive average accuracy of 97.96% in predicting future activities.
• The deep learning models proved effective in both forecasting IMU signals and classifying future activities.

Conclusions:

• The proposed HAP system effectively predicts future human activities using forecasted IMU data.
• This approach offers a significant advancement over traditional HAR systems by enabling proactive interventions.
• The high accuracy and correlation suggest strong potential for real-world applications in healthcare and safety.