Multi-Feature Transformer-Based Learning for Continuous Human Motion Recognition with High Similarity Using mmWave...

Area of Science:

• * Utilizes Doppler-radar-based continuous human motion recognition (CHMR) for enhanced privacy and personal security applications.

Background:

• * Existing CHMR methods using mmWave FMCW Radar struggle with recognizing continuous human motions that exhibit high similarity.
• * This limitation necessitates advancements in radar-based human motion analysis.

Purpose of the Study:

• * To propose a novel CHMR algorithm, termed CHMR-HS, designed to overcome the high similarity problem in motion recognition.
• * To investigate the effectiveness of spectral-temporal multi-features (2D and 3D) within a modified Transformer learning model.

Main Methods:

• * Employs background denoising and target extraction to clarify FMCW radar images.
• * Extracts 2D features (Doppler, range, angle signatures) and 3D features (range-Doppler-time, range-angle-time signatures).
• * Utilizes a modified Transformer-encoder and a re-designed Transformer-sequential-decoder model for distinguishing high-similarity motions.

Main Results:

• * The CHMR-HS scheme achieved accuracies of 95.2% with 3D features and 94.5% with 2D features.
• * Simulation results demonstrate the proposed CHMR-HS scheme's superiority over existing CHMR methods.

Conclusions:

• * The developed CHMR-HS algorithm effectively addresses the challenge of high similarity in continuous human motion recognition.
• * The modified Transformer-based approach shows significant advantages for mmWave FMCW Radar-based human motion analysis.