Enhanced In-Home Human Activity Recognition Using Multimodal Sensing and Spatiotemporal Machine Learning Architecture

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Human Activity Recognition (HAR) is crucial for smart environments.
• Existing HAR systems often struggle with complex temporal dynamics and multimodal data integration.
• Wearable sensors and Real-Time Location Systems (RTLS) offer rich data for HAR.

Purpose of the Study:

• To develop an enhanced HAR framework utilizing advanced machine learning models.
• To leverage multimodal sensor data, specifically from wearable wristbands and RTLS.
• To improve the accuracy and robustness of human activity detection in home environments.

Main Methods:

• Developed a spatiotemporal machine learning model combining Convolutional Neural Networks (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and Neural Structured Learning (NSL).
• Employed sensor fusion techniques, integrating data from Inertial Measurement Units (IMU) and RTLS.
• Compared the proposed model against traditional machine learning baselines (e.g., Random Forest, Support Vector Machines).

Main Results:

• The proposed spatiotemporal model significantly outperformed traditional machine learning baseline models.
• Sensor fusion of IMU and RTLS data achieved 86.21% accuracy and 87.40% F1 score for routine daily activities.
• Individual sensor modalities showed lower performance compared to the fused approach.

Conclusions:

• The enhanced HAR framework demonstrates superior performance in recognizing human activities.
• Sensor fusion is highly effective for improving HAR accuracy in complex environments.
• The developed system has significant potential for applications in elderly care, smart homes, and healthcare monitoring.