Analysing Headway Spacing and Calculating Passenger Car Equivalent Values Using Computer Vision and International Dataset

Abstract

Accurate traffic flow data are crucial for effective transportation planning and management. Different vehicle types impact traffic flow variably, requiring distinct passenger car equivalency (PCE) factors for calculating intersection and road capacity. Headway and spacing data are essential to assess traffic density and service level. Conventional data collection methods are time-consuming and often inaccurate. Unlike existing studies, this study employed computer vision to measure mixed traffic stream volume in terms of passenger car equivalent and collect headway-spacing data with high accuracy. The vehicle detection and counting procedures provide the mandatory infrastructure for measuring mixed traffic stream volume and collecting headway and spacing data. Novel approaches were introduced to gather comprehensive traffic data, including passenger car equivalent values, headway, spacing, flow rate, vehicle speed and traffic volume, using a single system. A custom and comprehensive international dataset was collected to analyse these approaches. Our trained model achieved a mean average precision (mAP) of 97.4%, with accuracies of 95% for headway, 93% for spacing and 99% for PCE values. The dataset can be downloaded at https://github.com/burak-celik/atavehicledataset.