Human motion recognition based on feature fusion and residual networks.

Area of Science:

• Signal Processing
• Machine Learning
• Human-Computer Interaction

Background:

• Single-feature reliance in human motion detection leads to suboptimal recognition accuracy.
• Existing methods often struggle with precise feature extraction and classification.
• Need for advanced techniques to integrate diverse data streams for robust motion analysis.

Purpose of the Study:

• To develop a novel human motion detection system with improved recognition accuracy.
• To integrate Frequency Modulated Continuous Wave (FMCW) radar data with a refined Residual Network (ResNet) architecture.
• To explore the efficacy of a dual-channel fusion approach for motion feature extraction.

Main Methods:

• Captured human motion echo signals using FMCW radar.
• Preprocessed signals and applied 2D Fourier transform to generate Range-time Maps (RTM) and Doppler-time Maps (DTM).
• Engineered a dual-channel fusion residual network by upgrading ResNet18 with Inception V1 modules and integrating the Convolutional Block Attention Module (CBAM).

Main Results:

• The proposed dual-channel fusion network effectively recognized and classified human motions.
• Achieved a 1-4% enhancement in recognition accuracy compared to single-feature domain recognition methods.
• Demonstrated robust recognition capabilities through empirical validation.

Conclusions:

• The integration of FMCW radar and a dual-channel fusion ResNet significantly improves human motion detection accuracy.
• The CBAM-enhanced ResNet architecture offers superior feature extraction and classification performance.
• This approach provides a promising solution for accurate and reliable human motion recognition.