Automatic Vehicle Counting in Complex Traffic Zones Using Computer Vision: A Case Study in the City of Huancayo

Abstract

The analysis of road traffic is fundamental for obtaining reliable information about itself. This paper presents Automatic Smart Vehicle Counting System – Intelligent System (CONTEV-SYS), a framework of a vehicle counting system developed in Python, employing YOLO v8n and the OpenCV library. A mixed method approach was adopted. Initially, a survey in the Likert scale was applied to 40 specialist identified the most extensive processing methods, errors in the data, and the impact in the weather as critical barriers of the traditional manual traffic counting system. Later on, a quantitative validation was carried out in nine vial scenarios, where different circulation patterns are presented in the city of Huancayo, comparing the system with a careful and accurate manual counting system. The results show that the CONTEX-SYS system has an average estimation efficiency of 89.40%, a Mean Absolute Percentage Error (MAPE) of 10.60%, reaching a maximum of 96% in scenario number 5, with a conservative behavior, that is, reducing the false positives in the severe occlusion case (R² = 0.921). The system attain a high precision in the vehicle classification, standing out the identification of heavy vehicles. In terms of operation, it saved 49.49% processing time, ensuring that the analysis was carried out in strict real-time. Finally, its implementation results to be highly promising in the field of civil engineering, particularly in the specialties of transportation and traffic.