Mask Recognition and Detection Method Based on Improved Yolov3
Author:Yijian Zhu,Shengting Sun, Panbo He Author Unit: UNICOM(SHANGHAI) INDUSTRY INTERNET CO.,LTD, Shanghai, 200050, China
Abstract：It is a very important and challenging task to detect the masks in image or video data during epidemic surveillance. The difficulty of this work lies in the accurate positioning and recognition of the relatively small proportion of masks in complex environment. In order to solve these problems, the DL-YoloV3 ,which modified the popular YoloV3 algorithm, combined with DIou loss function, improved YoloV3 and realized mask recognition in video surveillance. In this paper, the hardware environment of single card GPU NVIDIA Quadro p4000 and processor i7-9700k is tested. The experimental results show that the accuracy of the improved yolov3 model is significantly improved without reducing the speed. The average accuracy of 94.72％ is obtained on 8565 self-built data sets.
