Mask detector_Access control system with or without masks

//2021.06.11write

//However, if plagiarism is detected, graduation is canceled; Therefore, we leave it private and disclose it after the examination results are released.

//Xitron🍊

 

 

 

 

Mask detector_Access control system with or without masks

 

 This refers to the situation in which coronavirus infection-19, the infection of SARS-CoV-2 first identified in Wuhan, Hubei Province, China in December 2019, is prevalent around the world. The World Health Organization declared an international public health emergency in January 2020 and upgraded it to a pandemic in March. As the increase in the number of coronavirus infections-19 has increased, inconveniences in real life have increased one by one, and as a representative example, it has been emphasized as mandatory for people to wear masks.

 

 Accordingly, the capstone coronavirus infection - 19 focus on the project. Coronavirus infection - 19 will prepare a project to help in prevention. Name of the 'masks, detection, access control system' and wear masks to wear a mask in government emphasis. Detection system is a deep running mask for this situation from in front of the gate of the building using a camera and the people who are the mask worn a mask wearer and a man not to alarm on.

 

 The primary objective of the project first, people entering or leaving the building of the masks worn in real time with masks to give notification in accordance with or without.Second, cctv, unfastened and the overall number of people wearing masks in dense in the form of space in real time a precise count in real time by changing the current situation so that it can be found.This is highly, such as classrooms, subways, wearing masks in a system can be monitored in real time, can be found for unfastened. we need project, Wearing a mask has become mandatory at this time due to the coronavirus infection-19. However, many visitors visit nearby supermarkets and convenience stores without wearing masks. There is a risk of infection due to the number of people who visit indoor buildings without wearing masks due to the reason of "Wait." If there is a system that can detect masks, access to buildings can be controlled in a simple way.

 I had a market analysis.Non-nuclear level, the SARS-related corona virus infection to 19 related to the system created by the thought of.Three points of market analysis in process.The first, changes in the market for user speed and convenience at the time to meet the validity shall be designed to meet the needs.We'll have to meet and simplicity and accessibility should be available with a simple manipulation.Second, some building control by checking whether or not to wear masks when they were so many places, but most still not fully in control settings.'automated preventive measures that identify the system over temperature and inpeullaep's awareness of the mask michagyongja' The third, commercialization of product analysis exists.The system is artificial intelligence based on the environmental recognition and more than the mask michagyongja chomiljip cheonja possible identification rate, and universal use.Performance is 98 % and unfastened detection rate is 98 % or more, and respiratory system protection status judgments (facial) object detection speed is greater than 5 frames per second, and the same time, less than 0.3 degrees, body temperature measurement deviation is more than 20 the number of real-time data reflected with key performance such as speed is less than 10 seconds.But these differentiated products and the respiratory system protection for alarm and masks worn in real time ⦁ detected unfastened and count to the total number of people.

 As a result, a system that detects whether a mask is worn and gives an alarm is produced. This clearly has a lot of risk factors, but there is also a demand for notification functions for those who do not wear masks, and we judged that it is a good thing to be able to quickly recognize changing situations in real time. Also, I thought it was easy to install, so it had a commercial value. The system includes semi-automatic data labeling, model learning, and GPU code generation for real-time inference. Furthermore, the first expected effect is to check the presence of masks for the number of people entering the building in a simple way. Secondly, all personnel can abide by the mandatory administrative measures of masks. Third, people who wear and don't wear masks can be identified in real time.

 It divided development categories into 5 categories based on business analysis, design, development, test, and results. Details were set by requirements analysis, similar system analysis, system structure design, system function design, Dataset collection, data labeling and learning, mask-wearing notification function, error testing, integrated testing, final report preparation and presentation.

 I will explain the system design. The system is configured to allow images to be taken via webcam and then the Yolo model to determine whether to wear a mask. There are two main requirements for system functionality. The first 'mask presence detection' is labeled after mask detection, which utilizes Open CV and deep learning YOLO models to determine mask presence. The second "notification of wearing or not wearing a mask" function is to notify the wearing of a mask after determining whether it is not wearing or not, but the notification function is implemented by judging whether wearing a mask or not.(Sound notification is made by voice that you should wear a mask when you wear a mask and wear a mask when you don't.) The development environment was Library with Darknet, OpenCV, Language with Python, and Hardware with Desktop. The scope of development is H/W real-time imaging through USB webcam, S/W is first, mask presence is second, mask is correctly worn (including mask over nose) third, mask wearing and not wearing alarm sound is *when wearing mask: certified. *If you don't wear a mask: Please wear a mask.) Third, count the number of mask wearers and non-wearers in the real-time video.

 Core technology of producing object detection darknet and yolo with (you only look once) object recognition algorithms.Before and now learning to use the technology of the above aduwoyaal know you have.In order to to run the yolo darknet yolo of the files and weights cfg files needed.Additionally, object to learn some parameter files, weight and image files are learning in order to learn and learning objects that are completed, weighted to recognize the file is created.First, learning prepared for the datasets. ①dataset search provided by the Open and set of images. ② annotation tool the labeling the data used. ③the data type yolo the change in the form. ④class train, files and test names, valid the path of the images are a list of file exists. ⑤be Completed dataset.

set	Number of Images	Number of people wearing masks	Number of people wearing no masks
total	1020	3828	1073
train	700	3047	868
valid	200	278	49
test	120	503	156

 *train/val/test the ratio of images is usually divided by 7:2:1 or 6:2:2:2.

 Second, we do data learning. Images to be learned and cfg files and learned weight files are required. ①Modify the contents of the yolov3.cfg file to proceed with the study. ②Create an obj.names file. ③Create the file obj.data. ④Learning types are divided into batch, ubdivisions (mini-batch), max_batch (set the number of iterations) and epochs. ⑤In our case, we install cuda and cuDNN for learning with GPUs in Google colab environments. ⑥Install and compile the darknet. ⑦Run the darknet. ⑧Transfer files for learning and conduct learning in Google colab environment. ⑨After learning, weights (learning models) are stored in the backup folder of Google Drive.(Select the last weight (learning model) file stored because it has the lowest loss rate)

// ㅠㅠ sry, I don't know why I cut by picture

 

*The average loss rate is 0.7

 Consequently, the performance is shown in the table below.

Model	Training Set	Verification Set	test Set	FPS*
YOLOv3	99.75% 698/700	90.18% 180/200	87.16% 104/120	10FPS

*Standard cpu

 In this way, the Capstone design project "Access control system with or without masks" was completed.

 Thank you very much.💛