Gesture Recognition with Residual LSTM Attention Using Millimeter-Wave Radar

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Millimeter-wave (mmWave) radar enables non-contact gesture recognition.
• Existing neural network approaches face challenges with multi-feature complexity and single-feature performance limitations.

Purpose of the Study:

• To develop an advanced gesture recognition algorithm for mmWave radar systems.
• To overcome the limitations of current methods in handling complex and sparse gesture data.

Main Methods:

• A novel algorithm named ResNet Long Short-Term Memory with Attention (RLA) was proposed.
• Signal processing involved extracting range and velocity features to create range-Doppler maps.
• Network architecture integrated residual networks with channel and spatial attention modules, alongside a Long Short-Term Memory network for temporal feature processing.

Main Results:

• The RLA algorithm demonstrated enhanced focus on critical gesture features by employing a residual attention mechanism.
• High recognition accuracy was achieved even with single-feature inputs due to the Long Short-Term Memory component.
• Experimental results confirmed superior recognition performance compared to existing methods.

Conclusions:

• The proposed RLA algorithm offers a robust solution for mmWave radar-based gesture recognition.
• The integration of attention mechanisms and Long Short-Term Memory networks significantly improves accuracy and feature utilization.
• This approach effectively addresses the challenges of data complexity and performance variability in gesture recognition.