Performance Analysis of Vehicle Counting and Classification Using YOLO at a Four-Way Intersection Based on Indonesian Road Capacity Manual

This study introduces a vehicle volume analysis system utilizing video analytics technology to improve traffic counting accuracy at intersections using CRISP-DM methodology. By leveraging the YOLO (You Only Look Once) algorithm, it can detect, classify, and count vehicles using footage from public CCTV cameras based on Indonesian Road Capacity Manual. This research focuses on a four-way intersection in Sedayu, Yogyakarta, Indonesia, and involves preparing a dataset by extracting and labeling video frames. This research train and evaluate several YOLO models-YOLOv5s, YOLOv7, YOLOv8s, and Gelan-c from YOLOv9-to assess their effectiveness in real-world traffic scenarios. The results demonstrate that YOLOv8s delivers superior performance compared to the other versions, achieving a precision of 0.971 and a recall of 0.988 across all vehicle classes. This high accuracy highlights the potential of YOLOv8s for improving traffic volume counting at busy intersections. However, challenges remain, including false positives, double detections, and misclassifications, which tend to occur under certain traffic conditions, such as low visibility or dense vehicle clustering. Despite these issues, YOLOv8s shows promise for enhancing the efficiency and precision of traffic analysis systems. The contribution of this paper is to support the evaluation of intersection performance based on turn movement count in accordance with the Indonesian Road Capacity Manual, where currently the evaluation is still carried out manually. Future research will focus on addressing these challenges by refining the detection process and incorporating additional techniques, such as advanced post-processing methods and model enhancements, to increase robustness in various traffic environments.