Target Recognition Based on Millimeter-Wave-Sensed Point Cloud Using PointNet++ Model.

Area of Science:

• Biometrics
• Human-Computer Interaction
• Radar Signal Processing

Background:

• Gait recognition systems often struggle with performance degradation due to variations in arm movements.
• Collecting comprehensive arm motion data for training is a significant challenge in gait analysis.
• Millimeter-wave radar offers a non-intrusive method for capturing human motion data.

Purpose of the Study:

• To investigate the impact of diverse arm movements on radar-based gait recognition accuracy.
• To propose and validate a novel gait recognition method that utilizes only lower limb motion data.
• To enhance the robustness and accuracy of gait recognition systems against variations in arm gestures.

Main Methods:

• Collected gait data using millimeter-wave radar from 11 volunteers under various arm motion conditions.
• Developed a gait recognition approach based on extracted lower limb motion data using the PointNet++ model.
• Conducted comparative experiments evaluating the proposed method against a traditional method using all limb data.

Main Results:

• The proposed method using only lower limb data achieved a 22.9% improvement in accuracy compared to methods using all limb data.
• The new approach demonstrated enhanced feature extraction, accelerated convergence, and higher accuracy, particularly with limited samples (up to 96.9%).
• The method showed significantly superior performance and robustness in cases with unseen arm movements.

Conclusions:

• Excluding arm movements from radar-based gait recognition significantly improves system accuracy and robustness.
• The proposed lower limb-focused method is effective for mitigating interference from diverse arm motions.
• This approach offers a more reliable and efficient solution for gait recognition, especially in real-world scenarios with unpredictable arm movements.