Research On Mask Wearing Detection In Public Places Based On Deep Learning

Requiring everyone to wear mask correctly in public places during the pandemic is an effective measure to protect people’s safety and health.If intelligent supervision of mask wearing in public places can be realized,it will significantly reduce staff’s workload and improve regulatory efficiency.With a powerful feature learning ability,deep learning can solve complicated problems of pattern recognition.Based on deep learning,this project conducts research on the detection of mask wearing in public places,and implements a mask wearing detection system.The main work and its innovations are as follows:1)Considering there is no public and unified data set for mask wearing detection at complex scenarios,the project screened out the high-quality images in the RMFD data set and the WIDER FACE data set,and then supplemented the images through camera image capturing and web crawler.Finally,the project set up a mask-wearing detection data set containing 5515 images through manual annotation and expanded the image by online data enhancement,which improved the generalization of the model.2)In view of the problem that the detection accuracy of mask wearing is not high due to the complex scenes such as crowded,obscured,side-faced and small-scale in public places,the project made improvements to the YOLOv3(You Only Look Once version 3)algorithm featuring well-performed accuracy and speed.The first step is to improve the feature pyramid network of YOLOv3 algorithm through jump connection and Location Feature Enhancement(LFE)including channel attention.The abundant location information of low-level feature map is sent to the middle-level feature map and high-level feature map,which enhances the recognition to masks and faces.The second step is to improve the positioning accuracy of the algorithm by using the CIo U loss function instead of the mean square error loss function to perform frame regression.Besides mask wearing and non-wearing,the improved YOLOv3 algorithm also detects incorrect mask wearing.The experimental conclusion indicated that the improved YOLOv3 algorithm raised 3.30 percentage points compared with the YOLOv3 algorithm by achieving a m AP of 86.96% on the self-made mask wearing data set.The result is also better than the mainstream algorithms such as Faster R-CNN,SSD300,DSSD321 and YOLOv4.And the detection speed of this algorithm reaches 39.2 frame/s,which meets the real-time detection requirements.3)With the improved YOLOv3 algorithm and the multi-target tracking algorithm Deep SORT,the project built a mask wearing detection system including an image acquisition module,a mask wearing detection module and a people-tracking module to further validate the detection performance of the algorithm in real-world environments.After testing,the system can accurately and quickly detect people who do not wear mask or who do not wear mask correctly,achieving the intelligent detection of mask wearing in public places.Figure 38;Table 13;Reference 85...