Position-Aware Indoor Human Activity Recognition Using Multisensors Embedded in Smartphones.

Area of Science:

• Computer Science
• Biomedical Engineering
• Artificial Intelligence

Background:

• Composite indoor human activity recognition is crucial for elderly health monitoring but presents significant challenges compared to recognizing individual movements.
• Existing methods often lack integration with location data, limiting their effectiveness in complex indoor environments.

Purpose of the Study:

• To propose a novel sensor-based human indoor activity recognition method that integrates indoor positioning for enhanced accuracy.
• To explore the feasibility of using multi-sensor data from smartphones for comprehensive activity recognition.

Main Methods:

• Utilized convolutional neural networks (CNNs) to extract spatial features from geomagnetic and ambient light sensors.
• Employed transform encoders to capture temporal motion dynamics from gyroscope and accelerometer data.
• Developed a multimodal feature fusion model incorporating location information.

Main Results:

• Established and utilized a novel multisensor indoor activity dataset collected using smartphones.
• Demonstrated the effectiveness of the proposed multimodal fusion model through extensive experiments.
• Achieved a 13.65% improvement in recognition accuracy compared to methods that do not incorporate location information.

Conclusions:

• The proposed sensor-based method effectively recognizes composite indoor human activities by integrating indoor positioning.
• Smartphone-based multimodal sensor fusion offers a promising and accurate solution for elderly health monitoring applications.