Pengembangan Sistem Deteksi Dan Klasifikasi Kendaraan Menggunakan Yolo Dan Api Real-Time Dengan Metode Rapid Application Development (Rad)

Advances in computer vision technology have enabled the implementation of automatic vehicle detection systems using CCTV cameras. However, the application of deep learning models in real-world conditions still faces challenges, such as misclassification between vehicle types, variations in lighting, and duplication of calculations due to a single vehicle being detected in multiple frames. This research aims to develop a YOLOv8-based vehicle detection, classification, and counting system optimized for real-world traffic conditions through a rule-based post-processing approach and the ByteTrack tracking algorithm.
The research methods include requirements analysis, system design, YOLOv8 implementation, rule-based post-processing, and ByteTrack integration for multi-object tracking. Post-processing utilizes bounding box geometric information, such as area, aspect ratio, and object position in the frame, to improve vehicle classification accuracy. The ByteTrack algorithm is used to ensure each vehicle is counted only once through a stable track ID mechanism.
Tests were conducted using three test videos with different environmental conditions, as well as real-time testing using live CCTV. The results showed that the application of rule-based post-processing significantly improved precision, recall, and F1-score compared to no post-processing. The ByteTrack integration also successfully reduced duplicate vehicle counts and maintained system stability in real-time testing. The developed system was deemed capable of accurate and stable vehicle detection, classification, and counting under real-world CCTV conditions.