Intelligent Real-Time Face-Mask Detection System with Hardware Acceleration for COVID-19 Mitigation.

Area of Science:

• Computer Vision
• Deep Learning
• Hardware Acceleration

Background:

• The COVID-19 pandemic highlighted the need for effective real-time face-mask detection systems.
• Existing solutions often lack efficiency on embedded platforms, limiting widespread deployment.

Purpose of the Study:

• To propose and implement a dedicated hardware-accelerated deep learning system for real-time face-mask detection.
• To benchmark the performance of the proposed system on various embedded platforms.

Main Methods:

• Development of a novel deep learning model (MaskDetect) for face-mask detection.
• Independent quantization and optimization of MaskDetect for Raspberry Pi 4B (with Google Coral USB TPU or Intel Neural Compute Stick 2 VPU) and NVIDIA Jetson Nano.
• Ablation study comparing MaskDetect against VGG16, ResNet-50V2, and InceptionV3 on embedded platforms.

Main Results:

• MaskDetect achieved over 94% accuracy on most embedded platforms, with Coral achieving nearly 90%.
• The NVIDIA Jetson Nano implementation demonstrated the best balance of detection accuracy (94.2%) and inference speed (FPS).
• Jetson Nano offered twice the FPS compared to desktop hardware counterparts.

Conclusions:

• The proposed MaskDetect system provides efficient and accurate real-time face-mask detection.
• NVIDIA Jetson Nano is identified as the optimal embedded platform for this application due to its performance and cost-effectiveness.