Vision-Based Vehicle Classification and Load Estimation for Bridge Monitoring

This paper presents an intelligent system for vehicle classification and load estimation on bridges by integrating computer vision and Internet of Things (IoT) technologies. Accurate traffic load estimation is crucial for assessing bridge fatigue and predicting structural lifespan, as both factors determined by the stress range and loading frequency experienced by critical structural components. The proposed system facilitates real-time bridge monitoring without requiring physical weighing infrastructure. An ultrasonic sensor connected to an ESP32 microcontroller is used to detect passing vehicles and subsequently triggers a Raspberry Pi to capture video streams, which are then processed by the YOLOv8n object detection algorithm for vehicle classification. Load estimation is performed indirectly through a combination of bounding box dimension analysis, polygonal region-ofinterest filtering, one-dimensional Kalman filtering for smoothing, and dimensional feature matching with a predefined vehicle database. The classification and estimation results are transmitted via REST API and stored in a PostgreSQL database, while a web-based monitoring interface displays realtime data for infrastructure stakeholders. Experimental evaluations show that the system operates reliably with a vehicle detection precision of 91.2 % (as achieved on the validation dataset) and demonstrates consistent real-time classification performance in practical bridge monitoring scenarios. For vehicle load estimation, the system achieves a categorical accuracy of approximately 85 % in assigning vehicles to visual load categories based solely on external dimensions.