[1]毛晓波,徐向阳,李楠,等.基于改进SSD和Jetson Nano的口罩佩戴检测门禁系统[J].郑州大学学报(工学版),2021,42(6):87-94.[doi:10.13705/j.issn.1671-6833.2021.06.002]
　Mao Xiaobo,Xu Xiangyang,Li Nan,et al.A Mask Wear Detection Access Control System Based on Improved SSD and Jetson Nano[J].Journal of Zhengzhou University (Engineering Science),2021,42(6):87-94.[doi:10.13705/j.issn.1671-6833.2021.06.002]

点击复制

基于改进SSD和Jetson Nano的口罩佩戴检测门禁系统()

分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:

42

期数:

2021年6期

页码:

87-94

栏目:

出版日期:

2021-11-10

文章信息/Info

Title:

A Mask Wear Detection Access Control System Based on Improved SSD and Jetson Nano

作者:

毛晓波¹,徐向阳¹,李楠¹,魏刘倩¹,刘玉玺¹,董梦超¹,焦淼鑫²

1．郑州大学 电气工程学院，河南 郑州 450001; 2．郑州大学 机械与动力工程学院，河南 郑州 450001

Author(s):

Mao Xiaobo¹; Xu Xiangyang¹; Li Nan¹; Wei Liuqian¹; Liu Yuxi¹; Dong Mengchao¹; Jiao Miaoxin²;

1．School of Electrical Engineering，Zhengzhou University，Zhengzhou 450001，China; 2．School of Mechanical and Power Engineering，Zhengzhou University，Zhengzhou 450001，China

关键词:

口罩佩戴检测; 门禁系统; 目标检测SSD; Jetson Nano; MobileNet-V3

Keywords:

mask wear detection;  access control system;  object detection algorithm SSD;  Jetson Nano;  MobileNet-V3

DOI:

10.13705/j.issn.1671-6833.2021.06.002

文献标志码:

A

摘要:

为了减少疫情期间人们未佩戴口罩造成的交叉感染概率，设计一款基于改进的 SSD 和 Jetson Nano 的口罩佩戴检测门禁系统，以快速检测进出口行人是否佩戴口罩，控制闸机的开合。首先，从MAFA 和 WIDEＲ FACE 这 2 个数据集中抽取适合用于该系统的训练图片，其中 6 000 张作为训练集，2 000张作为测试集; 其次，利用随机色相、饱和度等像素级变换和随机扩展、随机裁剪等几何级变换，对数据集中的小目标进行数据增强，使数据集更加多样，增强该检测网络的泛化能力; 再次，将原始 SSD 的VGG 特征提取网络替换为 MobileNet-V3，利用其深度可分离卷积的速度优势，以及计算量较小的 HSwish激活函数、轻量化的注意力机制等优化策略，加速检测、提高精度; 最后，将该检测网络移植到计算能力有限的人工智能边缘计算设备 Jetson Nano 上，加装高清显示器，并设计可折叠的平行四边形挡板，选择合适的外围设备，构成了一个具有防疫价值的快速检测公共场所进出口行人是否佩戴口罩的多功

能门禁系统。在该嵌入式设备上的测试结果表明: 以 MobilNet-V3 为特征提取网络的目标检测算法SSD，取得了 78%的 MAP，FPS 为 12，与以 VGG 为特征提取网络的原始 SSD 算法( FPS 为 2) 相比，检测速度是原始 SSD 算法的 6 倍。该系统在保证实时性的同时也兼顾了检测精度，达到了精度和速度的平衡。

Abstract:

In order to reduce the chance of cross infection caused by people not wearing masks during the epidemic, a mask wearing detection access control system based on improved SSD and Jetson Nano is designed to quickly detect whether pedestrians at the entrance and exit wear masks and control the opening and closing of the gate. Firstly, 6 000 training pictures suitable for the system are extracted from the two data sets of MAFA and WIDER FACE, which are used as the training set and 2 000 as the test set; Secondly, the pixel level transformations such as random rue and saturation and geometric level transformations such as random expansion and random clipping are used to enhance the small targets in the data set, so as to add more samples to the data set and enhance the generalization ability of the detection network; Thirdly, the VGG feature extraction network of the original SSD is replaced by MobileNet-V3, which makes use of its speed advantage of depth-wise separable convolution, as well as the optimization strategies such as H-Swish activation function with less computation and lightweight attention mechanism (squeeze and excite) to accelerate the detection and improve the accuracy. Finally, the detection network is transplanted to Jetson Nano, an artificial intelligence edge computing device with limited computing power, equipped with high-definition display, design a foldable parallelogram baffle, and select appropriate peripheral equipment to form a multi-functional access control system with epidemic prevention value to quickly detect whether pedestrians at the entrance and exit of public places wear masks. The test results on the embedded device are as follows: the target detection algorithm SSD with MobileNet-V3 as the feature extraction network obtains 78% MAP and FPS is 12. Compared with the original SSD algorithm with VGG as the feature extraction network (FPS is 2), the detection speed is increased five times. Facts have proved that the system not only ensures the real-time performance, but also takes into account the detection accuracy, so achieves the balance of accuracy and speed.

常用功能