Real-Time Vehicles Detection with YOLOv8
Chih‐Jer Lin, C. Alisdair Lee · 2024
Given the severe congestion during peak hours, various traffic violations in Taiwan, and the development of urban technology, autonomous driving and vehicles will inevitably become a trend in the future. To address these issues, this study utilizes YOLO (You Only Look Once) object detection for vehicle recognition. It conducts in-depth research on YOLOv5, YOLOv6, YOLOv7, and YOLOv8, exploring the differences in training results among these versions under the same conditions. To enhance training outcomes, in addition to using publicly available datasets, actual footage from dash cameras is utilized. Vehicles are manually labeled using Labelimg to improve the accuracy of training results. The trained models demonstrate that the average precision mAP (mean average precision) can reach up to 99.5%.