Evaluating YOLOv5 and YOLOv8: Advancements in Human Detection

Abstract

The YOLO (You Only Look Once) method is a state-of-the-art approach in real- time object detection, known for its high-speed image processing capabilities. Recently YOLO versions have differed in performance, particularly in terms of detection accuracy and computational efficiency. The objective of this study is to assess the effectiveness and performance of YOLOv5 and YOLOv8 in real-time human detection applications using the SEMMA (Sample, Explore, Modify, Model, and Assess) methodology also. The dataset was processed through the Roboflow platform, which facilitated both the dataset management and the labeling process. Roboflow's tools streamlined the annotation of images, ensuring consistent labeling for deep learning model training and evaluation. F1 score, recall score, and precision score are compared both YOLOv5 and YOLOv8 to evaluate the performance of these architectures. The result of the evaluations shows that the performance of the YOLOv8 is better than the YOLOv5 which, YOLOv5 achieved F1-score equal 0.5865 (58%), recall score equal 0.83 (83%), and precision score of 0.4535 (45%). Meanwhile, YOLOv8 demonstrated better performance, with F1-score of 0.7921 (79%), recall score of 0.8289 (82%), and precision score of 0.7585 (75%). Base on the evaluations, we concluded that the performance of the YOLOv8 model is greater than the YOLOv5 model for Precision, and F1-Score, while YOLOv5 has slightly better score on recall. The contribution of this study is going to implemented into Audio guidance for the blind’s prototype that have been developing in previous study.

Keywords

• yolo
• object detection
• human detection