Comparative study of YOLOv3 and YOLOv5's performances for real-time person detection

Abstract

Deep learning algorithms have recently gained traction in smart cities and societies, such as in healthcare, surveillance, and in a variety of artificial intelligence-based real-life applications. Person detection is a big challenging task in modern computer vision applications. Thus, human appearances can be difficult to judge, as there are major differences in human postures and appearances. This paper discusses the performances of YOLO algorithms, especially YOLOv3 and YOLOv5 for person detection as a tool to enhance the security of public places in smart cities. To evaluate the performances, a challenging dataset called Penn-Fudan is used. Experimental results reveal that YOLOv3 outperforms YOLOv5 in terms of speed. However, YOLOv5 had the best recognition accuracy.