People Counting by Dense WiFi MIMO Networks: Channel Features and Machine Learning Algorithms

Area of Science:

• Ambient Intelligence
• Wireless Sensing
• Machine Learning

Background:

• Subject counting systems are crucial for ambient intelligence applications like smart homes and retail.
• Existing methods often require dedicated sensors, limiting flexibility.

Purpose of the Study:

• To develop a passive subject counting system using unmodified WiFi infrastructure.
• To explore machine learning techniques for accurate subject detection and counting.

Main Methods:

• Utilized multi-dimensional channel features from WiFi signals to detect subject presence.
• Compared Bayesian and neural network models for subject discrimination and counting.
• Employed ensemble classification to combine diverse learning models and leverage space-frequency diversity.

Main Results:

• Ensemble classification significantly improved counting accuracy by combining multiple models.
• The system achieved 99% average accuracy in detecting up to five moving people in an indoor environment.
• Considered real-time computing and cloud migration for practical deployment.

Conclusions:

• Unmodified WiFi infrastructure can be repurposed as a flexible and accurate passive sensing system.
• Machine learning, particularly ensemble methods, enhances subject counting performance.
• The proposed WiFi-based sensing offers a cost-effective solution for ambient intelligence scenarios.