【Academic English Homework】（Score：98）An application of Computer Vision under epidemic: A Simple implementation of Mask Wearing Detection in Active Scene

Abstract

Computer Vision and artificial intelligence facilitate the control of the COVID-19 in many ways. This essay is about one of the simple applications of Computer Vision – mask wearing detection from basic theories to experiments, which shows what can artificial intelligence contribute under epidemic. And the research on the algorithm of mask wearing detection has important application value and practical significance for the current epidemic prevention and control and even the post-epidemic era.

Keywords

mask wearing detection; deep learning; computer vision; artificial intelligence; COVID-19; object detection

Introduction

1.Background

At present, the global COVID-19 epidemic is raging, wearing a mask has been proved to be the most effective means of epidemic prevention. How to detect whether a person wears a mask has gradually become a hot topic of research. Today, the epidemic has entered its third year, and from Delta to Omicron, the threat posed by the virus to humans has not abated. In order to assure civilians’ health and safety, the Chinese government repeatedly asked people to wear masks in scientifically in public, however, there are still many people offend the provisions on prevention and control of the epidemic situation. Hence, it is essential to check and remind pedestrians to wear masks in public places. Manual inspection is ineffective, slow, and prone to false detections and missed detections. This paper investigates the algorithm of mask wearing detection in active scene and the real-time test shows the validity of the model.

2.Current status of research in this field

When an epidemic breaks out, experts are interested in learning how to effectively use computer vision to determine whether a person is wearing a mask. The problem that algorithms should solve is the decline of crucial characteristic point after wearing a mask. Due to this, traditional object detection may get wrong answer while detection. Deep learning in the Computer Vision field shows its power and many effective detecting algorithms have put forward in the past three years.

3. Research objectives

The Active scene and Passive scene are the two main key contexts for the detection of mask-wearing problems. The former is utilized in one-by-one detection in applications like access control, while the latter is employed in crowded public settings to simultaneously detect a large number of people. The main top of this essay focuses on Active scene.

Figure 1-A comparative analysis of two application scenarios

4.Passage structure

After simply introduce and describe some terminologies in computer vision field, the essay will show a basic detecting method by make an experiment step by step and clarify the principle. Besides, the essay will draw a conclusion and then compare this algorithm with other programmers’ algorithms.

Materials and methods

1.Terminologies ^[1]

The most crucial representative network design in deep learning is convolutional neural networks (CNN). CNN is able to imitate the mechanism of biological vision system, fully learn the features of images, sounds, etc. In this way, CNN may gather many feature information and identify the detection set following deep learning.

Figure 2-Convolutional neural network component

2.Experiment ^[2]

Objectives: Hundreds of masked and unmasked image datasets.

Environment: jupyter notebook with conda-py3.7.6-tf2.0.

Steps:

1. Create a notebook in jupyter notebook, choose conda-py3.7.6-tf2.0 as kernel

Code to copy, paste and visit the data sets.

2. Image acquisition and preprocessing

Import related packages:

Get data:

Image preprocessing:

Show some samples:

3. Create CNN

Initialize and print:

[In]

[Out]

4. Train

[In]

[Out]

5. Model prediction

[In]

[Out]

Results

Figure 3-A person who don’t wear mask

The CNN we created accords with the theoretical model and successfully detect a person who don’t wear mask in our data sets.^[3]（From RMFD^[4]、AIZOO、MaskedFace-Net^[5]）

Discussion

1.Conclusion

We could infer from the statistics mentioned above that computer vision has a great capacity for object detection. The CNN that we built can effectively predict whether the characters in the new image will wear masks or not through deep learning, it can be concluded.

2.Comparision

Comparing with YOLOv4^[5] series model, this simple CNN only can detect in active scene and its detecting speed is much lower than YOLOv4. However, it is simpler to make and requires less storage space.

3.Meanings

This simple experiment demonstrates the fundamentals of deep learning framework and lays the groundwork for more intricate algorithms.

4.Limitations and Prospection

As can be seen in the following image, this basic framework cannot detect people wearing masks correctly in more complicated settings:

Therefore, I strongly hope that more programmers will dedicate themselves to the study of mask detection in the future, effectively solve the issue, and use the solution in practical settings.

References

[1] [3] [5]王欣然, 人脸口罩佩戴检测与规范佩戴识别算法的研究[D]. 北京：北京建筑大学，2022：1-18; 26-27; 23-25.

[2] 广东软件协会，关于 2022 年广东省人工智能开发员职业技能竞赛赛前培训和变更初赛形式的通知[CP] 广东：广东软件协会，2022[2022.11.13]. http://www.gdsia.org.cn/publicfiles/business/htmlfiles/gdsia/cmsmedia/document/2022/10/doc22425.pdf

[4] Wand Z, Wang G, Huang B, et al. Masked Face Recognition Dataset and Application[J]. arXiv preprint arXiv: 2003.09093, 2020.

[5] Cabani A, Hammoudi K, Benhabiles H, et al. MaskedFace-Net – A Dataset of Correctly/Incorrectly Masked Face Images in the context of COVID-19[J]. Smart Health, 2021, 19:100144.