Sistem Klasifikasi Dan Deteksi Kendaraan Otomatis Dengan Custom Dataset YOLOv8 (Studi Kasus: Kota Balikpapan)

Abstract

Vehicle counting surveys are still conducted manually by deploying surveyors in the field. This approach faces several challenges, including the need for high concentration, physically demanding nature of task, and the requirement for many surveyors, which are inherent limitations of manual data collection. A viable alternative is to fully adopt artificial intelligence. This study employs one branch of machine learning, namely deep learning, to design an automatic vehicle detection system utilizing the YOLOv8 algorithm. The dataset was developed from camera footage at an intersection by capturing images of each passing vehicle. From these images, 80% were used for training and the remaining 20% for testing. The analysis results indicate the system’s performance achieved accuracy rates ranging from 96.92% in the morning to 100% during the day, and from 91.43% to 100% at night. Furthermore, the F1-Score values ranged from 67% to 100% in daytime, and from 80% to 100% at night.