Optimization of Traffic Performance Analysis in Kendari City with Deep Learning

Abstrak

Urban traffic congestion in developing cities like Kendari requires rapid and accurate monitoring solutions. This research aims to design and implement an integrated traffic performance analysis system utilizing the YOLOv8 deep learning architecture. The system, named KILALIN, automates vehicle detection, classification, and tracking to calculate road capacity and saturation levels based on the PKJI 2023 standards. A comprehensive dataset of 1,606 annotated images was utilized, partitioned into training (57%), validation (29%), and testing (13.7%) subsets. The developed YOLOv8s model achieved high performance with a mean Average Precision (mAP@0.5) of 0.948, precision of 0.941, and recall of 0.935 across all vehicle classes. Functional validation through black-box testing confirmed the system's ability to process real-time video inputs under various conditions. Comparative results with manual surveys showed a 96% counting accuracy, indicating the system's robustness in quantified traffic flow estimation. Furthermore, the integration of automatic Passenger Car Equivalent (EMP) conversion allows for an immediate determination of the Degree of Saturation (DS) and Level of Service (LoS). These findings indicate that the YOLO-based traffic performance analysis system provides a reliable and efficient framework for urban traffic management, effectively replacing conventional manual survey methods while maintaining high technical standards.