Nur Aina Syafinaz, Muhamad Atfan (2022) Analysis Of Human Detection Method In Social Distancing Monitoring. College of Engineering, Universiti Malaysia Pahang Al-Sultan Abdullah.
|
Pdf
EA18088_NUR AINA SYAFINAZ_THESIS - Aina Syafinaz.pdf - Accepted Version Download (3MB) | Preview |
Abstract
Social distancing is a non-medical practice that helps slow down the transmission of viruses which is suggested by the World Health Organization (WHO). Social distancing means that whenever people want to socialize, they need to adhere minimum suggested social distance of 2 meters apart from the people around them. However, as every country has battled the spread of the virus for almost three years, social distancing practice now seems ignored by public people due for some reason such as they are in a rush. This study aims to analyze human detection using the deep learning method in various positions and to develop social distancing detection using the proposed method. Detection means to recognize the presence of something such as human or object. In the area of detection, several methods are being used either traditional or modern technology. The proposed method for human detection in social distancing monitoring is by using a deep learning algorithm which is You Only Look Once (YOLO) version 3 with custom datasets. In this method, the Euclidean distance formula is used for measuring the distance between two people with another additional computation to produce the output measured distance true to the real-life measurement. The datasets are collected from online sources which are from Kaggle, Unsplash, and iStock. There are about 403 images is used for this study. 282 images are used for the training process and 121 images are used for the testing process. There are another additional 300 customed datasets are used to test the human detection and social distancing detection using the same detection model. Google Colaboratory is used as a programming platform with the use of Python programming language. The custom datasets are being labeled first using Labelimg software before going through the training process. As a result, the current datasets for each position such as front view, back view, side view, and the crowd gives the result of human detection at 94.44%, 91.67%, 97.50%, and 88.89% respectively. Hence, the highest accuracy of human detection goes to the side view position with a percentage accuracy of 97.50%. Besides, using the same model for customed dataset gives overall percentage accuracy which is 98.00% in detecting humans. Next, the social distance detection is successfully developed by using the Euclidean distance formula by multiplying it by a calculated ratio of a pixel to centimetre which is 0.1978cm. Therefore, the measured distance is in centimetres representation. Hence, the detected social distancing result is also in the range of acceptable at ±0.3cm. For future planning, datasets need to be around thousand images for the training process to make the detection accuracies increase. This study concludes that the use of current proposed object detection algorithms can be used to detect humans and the Euclidean distance method is successfully implemented together with the object detection algorithm
| Item Type: | Undergraduates Project Papers |
|---|---|
| Additional Information: | SV: Dr. Rosdiyana Binti Samad |
| Uncontrolled Keywords: | Social distancing, You Only Look Once (YOLO) |
| Subjects: | T Technology > TA Engineering (General). Civil engineering (General) T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Faculty/Division: | College of Engineering |
| Depositing User: | Mr. Nik Ahmad Nasyrun Nik Abd Malik |
| Date Deposited: | 08 Jan 2024 07:55 |
| Last Modified: | 08 Jan 2024 07:55 |
| URI: | http://umpir.ump.edu.my/id/eprint/39897 |
| Download Statistic: | View Download Statistics |
Actions (login required)
 |
View Item |