嵌入式口罩佩戴检测系统研究与实现

Research and implementation of embedded mask wearing detection system

常态化疫情防控形势下,公共场合佩戴口罩可以有效降低交叉感染风险,针对口罩佩戴检测中的小目标检测困难以及实时性较差的问题,提出了基于嵌入式平台Jetson nano的口罩佩戴检测系统,通过增加YOLOv3-tiny的主干网络层深度,引入注意力机制以及TensorRT模块,提升了嵌入式系统口罩佩戴检测任务的精度和实时性,改进后的YOLOv3-tiny算法mAP值达到了87.5%,FPS为20.4,相较于改进前精度提升12.3%,帧率提升10.4 fps。

Under the situation of normalized epidemic prevention and control,wearing a mask in public can effectively reduce the risk of cross infection.In view of the difficulty in detecting small targets and poor real-time performance in mask wearing detection,a mask wearing detection system based on the embedded platform Jetson nano is proposed.By increasing the depth of the backbone network layer of YOLOv3-tiny,introducing the attention mechanism and the TensorRT module,the accuracy and real-time performance of the mask wearing detection task of the embedded system are improved.The improved YOLOv3-tiny algorithm has a mAP value of 87.5%and an FPS of 20.4.Compared with the previous improvement,the accuracy has increased by 12.3%and the frame rate has increased by 10.4 fps.