Deep Neural Network Based Performance Evaluation and Comparative Analysis of Human Detection in Crowded Images Using YOLO Models

Human detection is required for many important domains, including safety, people flow, health care, autonomous driving, and security monitoring. The You Only Look Once (YOLO) family of models has shown to be a promising approach for human detection. YOLO is a detection algorithm that identifies objects or human in images, videos or live feeds. This paper describes the analysis of different YOLO models such as YOLOv3, YOLOv4 and YOLOv5 for detecting the presence of human in the input images. Accuracy, precision and recall were the measuring methodologies used to assess how well the three algorithms were performed. Experimental analysis shows that YOLOv5 achieves an accuracy, precision and recall of 0.91, 0.90 and 0.90, respectively. Results confirm that YOLOv5 outperforms both YOLOv3 and YOLOv4 models.