Face mask detection of dense crowds using deep learning

• Main Author: Shangguan, Yuexi
• Other Authors: Yap Kim Hui
• Format: Thesis-Master by Coursework
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/159915
• Institution: Nanyang Technological University

Object detection is an important research direction in computer vision. Object detection algorithms are widely used in intelligent video surveillance, robot navigation, industrial inspection, aerospace, and other fields. The spread of the coronavirus disease (COVID-19) around the world has brought serious threats to the lives of people all over the world. The virus spreads in the air in the form of droplets and aerosols, and is highly infectious, especially in public places, posing a great threat to people’s health and safety. Wearing a mask in public places is the most effective measure to stop the spread of the virus. It is of practical significance to detect whether the crowd is wearing a mask or not. This project uses a deep learning-based object detection method to detect whether people wear masks. Through a detailed analysis of object detection, You Look Only Once-v5(YOLOv5) can perform well in the face mask detection scene in terms of real-time detection and detection accuracy. We adopt coordinate attention mechanism in this project to further improve the YOLOv5, called CA-YOLOv5. The attention mechanism can give more weights to important features and improve the utilization of important features. Finally, the Face Mask dataset and Factory dataset are used to evaluate the YOLOv5 with and without coordinate attention mechanism, and the evaluation metrics, like precision, recall and mean Average Precision (mAP) are shown on those two datasets. Moreover, this study com pares our method with other popular algorithms and tabulated their results. It can be seen from the table that the performance of CA-YOLOv5 is better than that of SSD, DSSD, Faster-RCNN and other algorithms of the YOLO series.