A semi-supervised Hidden Markov model-based activity monitoring system

Area of Science:

• Computer Science
• Biomedical Engineering
• Machine Learning

Background:

• Human activity classification systems often demand extensive training data, limiting their practical application.
• Developing robust models for diverse activities requires significant data collection and annotation.

Purpose of the Study:

• To propose a semi-supervised Hidden Markov Model (HMM) based system for human activity monitoring.
• To reduce the dependency on large, subject-specific training datasets for activity classification.
• To enable recognition of complex behaviors by temporally linking simple events.

Main Methods:

• Utilized a semi-supervised Hidden Markov Model (HMM) approach.
• Adapted a general HMM model to a specific subject's data.
• Employed two triaxial accelerometers for data acquisition.
• Developed a temporal linking mechanism for event-based behavior recognition.

Main Results:

• The system successfully adapts HMMs from general models to individual subjects.
• Demonstrated feasibility in identifying simple activities like sitting, standing, and walking.
• Showcased the ability to recognize more complex behaviors by sequencing basic events.
• Experimental results validate the proposed system's effectiveness.

Conclusions:

• The proposed semi-supervised HMM system effectively alleviates the need for large training datasets in human activity recognition.
• The system demonstrates a feasible approach for recognizing both simple and complex human activities using wearable sensors.
• This method offers a practical solution for personalized activity monitoring in various applications.