A Novel Energy-Efficient Approach for Human Activity Recognition

Area of Science:

• Computer Science
• Electrical Engineering
• Biomedical Engineering

Background:

• Mobile activity recognition systems (ARS) are crucial for various applications.
• High energy consumption of traditional ARS limits their practical deployment on mobile devices.
• Existing methods often require high sampling rates, leading to increased power demands.

Purpose of the Study:

• To propose a novel energy-efficient approach for mobile activity recognition.
• To achieve high recognition accuracy with significantly reduced energy consumption.
• To investigate the feasibility of using low sampling rates below the Nyquist limit for activity recognition.

Main Methods:

• Development of a novel classifier: hierarchical support vector machine and context-based classification (HSVMCC).
• Implementation of an energy-efficient ARS utilizing low sampling rates (e.g., 1 Hz).
• Experimental validation with data from 20 volunteers performing various activities.

Main Results:

• The proposed HSVMCC classifier achieves high accuracy (around 96.0%) even with low sampling rates.
• Using a 1 Hz sampling rate saves 17.3% and 59.6% of energy compared to 5 Hz and 50 Hz, respectively.
• The approach effectively reduces power consumption while maintaining high activity recognition performance.

Conclusions:

• The proposed low sampling rate approach offers a viable solution for energy-efficient mobile activity recognition.
• HSVMCC enables accurate activity detection under sub-Nyquist sampling conditions.
• This method significantly enhances the practicality of ARS for long-term mobile deployment by minimizing power usage.