Human movement detection and identification using pyroelectric infrared sensors

Area of Science:

• Sensor Technology
• Human-Computer Interaction
• Machine Learning

Background:

• Pyroelectric infrared (PIR) sensors are common for presence detection.
• PIR sensor analog output is influenced by movement dynamics, distance, and subject characteristics.
• Accurate human movement analysis requires advanced signal processing and classification.

Purpose of the Study:

• To empirically study human movement detection and identification using PIR sensors.
• To develop and evaluate a data collection module with orthogonally aligned PIR sensors.
• To assess the effectiveness of different feature sets and machine learning algorithms for human movement analysis.

Main Methods:

• Developed a data collection module with pairs of PIR sensors and modified Fresnel lenses.
• Deployed three PIR-based modules in a hallway for comprehensive monitoring.
• Collected data from eight subjects under varying movement conditions (direction, distance, speed).
• Utilized raw and reduced feature sets (amplitude, time to peaks, passage duration).
• Performed classification using instance-based learning and support vector machine algorithms.

Main Results:

• Achieved over 92% accuracy in classifying movement direction, speed, distance, and identifying subjects using raw data from single PIR sensors.
• Reached over 94% accuracy using a reduced feature set from multiple PIR sensor pairs.
• Demonstrated the efficacy of PIR sensor arrays for detailed human movement characterization.

Conclusions:

• PIR sensor arrays offer a robust solution for human movement detection and identification.
• Feature engineering and machine learning significantly enhance the accuracy of PIR-based human sensing.
• This approach has potential applications in security, surveillance, and intelligent environments.