Effect of Yolo-Based Vehicle Detection and Manuel Vehicle Classification Methods on Simulation-Based Intersection Analysis

In complicated intersections, for measuring intersection performance using simulation programs (PTV Vissim, Aimsun, Sidra, etc.), decomposing camera recording data procured for vehicle counting according to vehicle type and movement direction is laborious. In this study, a software application was developed using YOLO (You Only Look Once) object detection model to classify vehicle types and estimation the number of vehicles from images obtained with camera recordings. The system aims to provide high-accuracy results in significantly less time compared to manual visual counting methods. The proposed model achieved a detection accuracy of 91.8%. The software was tested on video data from the Sampi Intersection, and the results were validated by comparing them to manual counting. Based on the obtained data, performance analyses were conducted using the PTV Vissim software by evaluating vehicle delays, determining service levels, and making comparative assessments. As a result of the simulation analysis, the vehicle delays, which are 38.82 seconds in the manual count method and 37.36 seconds in the software counting method, are between 35.1-55 seconds and correspond to the service level value D. The results demonstrate that YOLO-based vehicle counting and classification systems can offer an efficient and reliable alternative for traffic monitoring and intersection analysis.