Efficient Markerless Motion Classification Using Radar.

Area of Science:

• Radar Signal Processing
• Biomedical Engineering
• Human Motion Analysis

Background:

• Marker-based motion capture systems are accurate but invasive and costly.
• Markerless motion classification is desired for broader applications.
• Micro-Doppler radar signals contain rich information about human body micro-motions.

Purpose of the Study:

• To develop a novel markerless motion classification method using radar.
• To extract effective features from micro-Doppler signals for motion analysis.
• To achieve high classification accuracy without physical markers.

Main Methods:

• Utilized 3D marker coordinates from motion capture to create basis functions for modeling micro-motions.
• Generated feature vectors via cross-correlation between radar signals and basis functions.
• Employed Principal Component Analysis (PCA) for feature compression and Nearest Neighbor for classification.

Main Results:

• Achieved nearly 100% classification accuracy with a compact feature set.
• Demonstrated accuracy even at high signal-to-noise ratios.
• Highlighted the importance of optimizing sampling duration and interval for efficiency.

Conclusions:

• The proposed radar-based method offers a highly accurate and efficient markerless solution for human motion classification.
• Basis functions derived from motion capture data effectively model human micro-motions from radar signals.
• Careful selection of sampling parameters is crucial for optimizing training data and computational performance.