Sensor positioning for a human activity recognition system using a double layer classifier

Area of Science:

• Biomedical Engineering
• Human Activity Recognition
• Wearable Technology

Background:

• Multi-sensor systems are effective for human gait activity recognition but are often cumbersome.
• Reducing sensor count is desirable for practical, user-friendly applications.

Purpose of the Study:

• To develop a single-sensor human gait activity recognition system with performance comparable to multi-sensor systems.
• To identify the optimal sensor placement for gait activity recognition.

Main Methods:

• A maximum relevance minimum redundancy (MRMR) feature selection method was used to determine the optimal sensor location.
• A random forest classifier was employed, with features selected using MRMR and a genetic algorithm.
• Algorithm modifications included a double-layer classifier and the addition of physical features to compensate for sensor reduction.

Main Results:

• The thigh was identified as the optimal sensor location for recognizing various gait activities.
• The modified single-sensor system achieved prediction accuracy comparable to multi-sensor systems.
• The double-layer classifier effectively discriminated between similar activities, and added physical features improved accuracy.

Conclusions:

• A single-sensor system can effectively perform human gait activity recognition.
• Optimized sensor placement and algorithmic enhancements are key to achieving high accuracy with reduced sensor count.
• This approach offers a more practical and less intrusive solution for gait analysis.