<![CDATA[This research discusses the implementation of a vehicle counting detection system on roadways utilizing the YOLO (You Only Look Once) algorithm, integrated with PyTorch's transfer learning and fine-tuning techniques. The study is motivated by the rapid increase in private vehicle ownership in Indonesia, which has heightened concerns regarding traffic congestion and accident risks. The primary objective of this research is to develop an efficient vehicle counting system based on Convolutional Neural Networks (CNN), designed to process images and videos. The methodology encompasses a literature review, system analysis, design, implementation, and evaluation. The system is built using YOLOv8, tailored with transfer learning to enhance object detection, focusing on cars. To track and count vehicles, the Centroid Tracker algorithm is employed. A dataset of 1,667 images was used, partitioned into training (1,488 images), validation (118 images), and testing (61 images) sets. The model achieved a detection accuracy of 97.92%, though minor detection errors were observed. In video-based testing, the system effectively detected and tracked vehicles, assigning unique IDs to individual cars. In conclusion, the YOLOv8-based model, combined with the Centroid Tracker algorithm, demonstrates strong performance in detecting and counting vehicles, offering potential contributions to traffic monitoring systems and providing a foundation for more sophisticated applications in future research.]]>
Copyrights © 2024
