基于YOLOv5s的人脸是否佩戴口罩检测

YOLOv5s-based detection of mask wearing

为了抗击新冠肺炎,加快复产复工的速度,本文提出了一种判断人脸是否佩戴口罩的方法。该方法可以通过摄像头捕获人脸,对人脸是否佩戴口罩进行检测,及时提醒人们带好口罩。使用预处理好的9 800张图片作为数据集,并对YOLOv5s算法进行离线训练,生成最终的模型;利用该模型对摄像头捕捉的画面进行分析,检测人脸是否佩戴有口罩。该算法在测试集上的精确率(precision)、召回率(recall)和平均精度(mAP)分别为78.1%、87%和53.5%,高于YOLOv3和YOLOv4检测算法的检测结果。

In order to combat the COVID-19 and accelerate the speed of resuming production and work,this paper proposes a method for detecting mask wearing. The method can capture face through cameras,detect whether a mask is on the face or not and promptly remind people to wear a mask. The method uses9 800 pre-processed pictures as a dataset,and trains YOLOv5 s network offline to generate the final model. Finally,the proposed algorithm uses the trained model to analyze and further to judge the face weather a mask is on it or not. The precision(A)、recall(R)and average precision(mAP)of this algorithm on test dataset are78.1%, 87% and53.5%,respectively. These results are higher than those of YOLOv3 and YOLOv4 detection algorithms.