Classification of Human Motions Using Micro-Doppler Radar in the Environments with Micro-Motion Interference

Area of Science:

• Radar Signal Processing
• Human Motion Analysis
• Machine Learning

Background:

• Micro-Doppler effect enables human motion classification.
• Non-target micro-motion interference degrades classification performance by creating cross-terms.
• Existing methods lack robustness against such interference.

Purpose of the Study:

• To develop a human motion classification system resilient to non-target micro-motion interference.
• To enhance the accuracy of human motion identification in cluttered wireless environments.
• To address the limitations of current methods in handling signal aliasing and cross-terms.

Main Methods:

• Utilized a continuous wave radar transceiver (USRP N210) in a low-frequency band.
• Applied Empirical Mode Decomposition and S-transform for interference removal and signal enhancement.
• Developed an Energy Aggregation method based on S-method to suppress cross-terms and noise.
• Extracted features and employed Bagged Trees for classifying four distinct human motions.

Main Results:

• Successfully removed non-target micro-motion interference.
• Improved time-frequency resolution of the raw radar signals.
• Suppressed cross-terms and background noise effectively.
• Achieved 97.3% classification accuracy in scenarios with interference.

Conclusions:

• The proposed system demonstrates high effectiveness in classifying human motions amidst non-target micro-motion interference.
• The combination of signal processing techniques and machine learning provides a robust solution.
• This work offers a significant advancement for reliable human motion detection in challenging environments.