Improving the public transport monitoring system using machine learning methods

This study focuses on developing software to address a practically important problem: providing public oversight of urban transport movement. This is particularly crucial in areas with limited GPS/GLONASS and cellular connectivity. The core of the system comprises video cameras located along the route and a computer vision system. This system detects the presence of buses or trolleybuses in the camera’s field of view, localizes them, and recognizes their route numbers. Using the freely available YOLOv11s object detector, a machine recognition accuracy of 96 % was achieved. This version of YOLO is resource-efficient, enabling the use of a standard personal computer to process multiple video streams. Route numbers were recognized using the open-source PaddleOCR library, achieving an accuracy of 82 %. The obtained results were compared with the bus schedule, and the data was posted via a Telegram bot. The research results aim to improve the convenience of urban public transport, reduce social tension, and provide residents and dispatching services with real-time information about deviations in urban transport operations.

urban public transport, computer vision, machine learning, YOLO, object detection, transport monitoring, automatic vehicle identification, image processing, PaddleOCR, transport analytics