Scalable automated traffic flow monitoring system for viaducts: A data-fusion approach for vehicle mass estimation

Efficient infrastructure management requires real-time traffic monitoring to address rising vehicle loads and enable timely maintenance. This study presents a data-fusion approach combining acceleration-based and vision-based systems to classify vehicles and estimate their weight over time. Validated on a real viaduct and benchmarked against a reference P-WIM (Pavement Weigh-In-Motion) system, the fused method significantly improves vehicle mass estimation compared to vision-only methods based on the YOLOv11 model. The combined system achieves individual mass estimation errors of 15% for light and 30% for heavy vehicles. However, cumulative traffic load estimation yields a total error of just 1.5%, as individual errors tend to sink over time. Relying solely on vision leads to a consistent 5% overestimation, underscoring the added value of acceleration data. While not a replacement for high-precision P-WIM systems, this scalable approach enables continuous, network-wide traffic load monitoring using existing infrastructure, supporting smarter maintenance planning and enhanced structural health management.