A Conceptual Real-Time Deep Learning Approach for Object Detection, Tracking and Monitoring Social Distance using Yolov5

Abstract

Objectives: To develop a computer vision-based model that can detect, track and recognize individuals for the purpose of measuring social distance in road traffic videos using surveillance cameras. Methods: The real-time traffic surveillance webcam dataset was applied to validate the model, with better performance metrics outperformed to those of comparable cuttingedge models such as RetinaNet, Faster R-CNN and SSD. Our proposed methodology utilized object detection methods to recognize individuals followed by multiple objects tracking approach to track identified individuals using detected bounding boxes. Our research shows that the conventional method is successful in detecting persons who violate social distances. Findings: Our finding shows that our proposed object detection model successfully recognizes human and those who violating the social distancing measurements. For the purpose of detecting social distance, develop a highly accurate detection technique. Our YOLOv5 with multiple objects tracking algorithm delivered great outcomes with appropriate Precision of 93%, Recall of 94%, F1-Score of 96% and mAP of 95% measures given by object categorization and localization to measure social distancing in real-time traffic surveillance videos. The YOLOv5 model’s results are then compared to many other prominent state-of-the-art models. Novelty: The YOLOv5 and MOTSORT is appropriate for finding whether people are maintaining social distancing or not, intended to identify, monitor and track those who are not following or violating social distances with overall accuracy and efficiency. Keywords: Object Detection; Deep Learning; Social distance monitoring; multiple objects tracking; YOLOv5