Outdoor Social Distancing Detector Using YOLO Algorithm

Yong, Chee Yeh (2022) Outdoor Social Distancing Detector Using YOLO Algorithm. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text Yong Chee Yeh.pdf Restricted to Registered users only Download (2MB)

Abstract

The outbreak of coronavirus disease 2019 (COVID-19) with deadly spread has brought about a global crisis. One of the effective approaches to fight against this pandemic is to ensure social distancing. Inspired by this concept, this paper researched to develop a deep learning-based framework for Outdoor Social Distancing Detector Using Yolov3 Algorithm which is used to monitor the social distance automatically using surveillance video. The captured video will first be extracted into images and further pre-processed to extract the image feature. Object detection is then applied for pedestrian detection. This system is mainly applied with YOLOv3 for object detection to measure the distance between people. The generated result will be further compared with the performance of the YOLOv4 algorithm in terms of speed and accuracy. Several approaches were used to analyze the performance. The selected algorithms are trained and tested with Google’s Open Images Dataset V6+ Extensions dataset and their performance is evaluated. Few hyperparameters are tuned to optimize the performance of the selected algorithm such as the “Early-Stopping” technique to avoid overfitting, transfer learning to speed up the training process as well as data augment technique. Software such as Visual Studio, Python extension, tensor flow, OpenCV-Python are used to carry out the training and testing. Based on the results shows that Yolov4 able to achieved 59.69% of mAP which 17.59% higher than YOLOv3, while for the FPS, YOLOv4 was 180.2% higher than YOLOv3 where YOLOv4 able to reach 5.38fps while YOLOv3 only 1.92fps. As a result, YOLOv4 perform better and more suitable to be selected as the algorithm for the social distancing detection system.