Human activity recognition using magnetic induction-based motion signals and deep recurrent neural networks

Area of Science:

• Biomedical Engineering
• Computer Science
• Signal Processing

Background:

• Human physical activity recognition is crucial for healthcare, rehabilitation, athletics, and senior monitoring.
• Existing wireless sensor networks face challenges in accuracy, power consumption, cost, and computational complexity, especially in lossy environments like the human body.

Purpose of the Study:

• To introduce an innovative wireless system for human activity recognition that addresses the trade-offs between power consumption, cost, computational complexity, and accuracy.
• To develop a sensor-based system capable of operating effectively in and around the human body.

Main Methods:

• Development of a novel wireless system utilizing magnetic induction.
• Integration of the magnetic induction system with machine learning techniques.
• Evaluation using synthesized datasets, laboratory measurements, and deep recurrent neural networks.

Main Results:

• The magnetic induction system successfully detects a wide range of human motions.
• The approach demonstrates effectiveness in tackling the inherent challenges of sensor-based activity recognition in lossy environments.
• Successful evaluation across diverse datasets and advanced neural network models.

Conclusions:

• The proposed magnetic induction wireless system offers a viable solution for human activity recognition.
• This innovative approach provides a promising trade-off among key system constraints for practical applications.
• Further development and application in real-world scenarios are warranted.