Sistema inteligente de gestión de tráfico mediante visión por computadora y Python para la identificación y estimación de velocidad vehicular

Abstract

Vehicle speed estimation systems based on computer vision have gained significant attention as support tools for effective traffic control strategies. In this article, vehicle detection algorithms and object tracking techniques were employed using an image processing library. A road segment with excellent visibility and lighting conditions was selected, and a fixed camera was installed at different heights to ensure image quality. In the methodology, the Region of Interest (ROI) was defined by selecting two pairs of real-world points and projecting them onto the video frames. Subsequently, the pretrained YOLOv8 model, fi-ne-tuned with labeled images from RoboFlow, was used for vehicle detection and classification; tracking was performed with DeepSORT. Speed was calculated from the time interval during which vehicles crossed the virtual lines corresponding to the entry and exit of the ROI, converting pixel distances into meters through a geometric calibration factor. All the information was systematically stored in a MySQL database. Validation was carried out with a GPS speedometer, yielding an average deviation of ±3.16 km/h and a relative error of 4.74%. This approach provides a practical solution for measuring vehicle speed without the need for complex sensors, enabling effective traffic monitoring and opening ap-plications in intelligent transportation systems.