Machine Learning-Based Human Posture Identification from Point Cloud Data Acquisitioned by FMCW Millimetre-Wave Radar

Area of Science:

• Engineering
• Computer Science
• Biomedical Engineering

Background:

• Human posture recognition is vital for healthcare, human-computer interaction, and sports.
• Frequency-Modulated Continuous Wave (FMCW) millimetre-wave (MMW) radar offers robust capabilities for measuring human posture characteristics.
• Existing methods may be limited by environmental noise and target reflection strength.

Purpose of the Study:

• To demonstrate the measurement, classification, and identification of human posture characteristics using FMCW MMW radar.
• To generate point cloud data that enhances target reflection and reduces environmental noise.
• To evaluate multiple machine learning models for accurate posture classification.

Main Methods:

• Human posture data was collected using MMW radar sensors.
• Point cloud data was generated from reflected signals, incorporating dynamic and static features.
• Six machine learning models were applied for classification, with Kappa index used for evaluation.

Main Results:

• Point cloud generation effectively reduced environmental noise and strengthened target reflections.
• The multi-layer perceptron (MLP) model outperformed other classifiers.
• The Kappa index was successfully used to address data imbalance issues.

Conclusions:

• FMCW MMW radar is a viable technology for human posture recognition.
• MLP is the most promising machine learning model for classifying radar-based human posture data.
• The developed point cloud generation method enhances data quality for posture analysis.