Social distance monitoring of site workers for COVID-19 using context-guided data augmentation, deep learning, and homography transformation

Abstract Because of the COVID-19 pandemic, many industries have developed efforts to minimize COVID-19’s spread. For example, construction industry in Melbourne practices social distancing and downsizes number workers on job site. The surveillance system integrated with deep learning models has been extensively utilized enhance safety. However, such 2D-based approaches suffer from occlusions, may not be accurately detected under this circumstance. To end, paper proposes a novel context-guided data augmentation method models’ performance occlusions. can automatically augment images by adding occlusions objects. Using way, learn object’s features various occlusion scenarios. Later, is validated real-time violation detection system. Specifically, utilizes modified YOLOv4 model detect bounding boxes. Then, DeepSORT algorithm used track worker trajectories. Finally, homography transformation calculate distance between each frame. revealed robust results using method, promising indicate that well support health during COVID-19.