Automating Traffic Monitoring with SHM Sensor Networks via Vision-Supervised Deep Learning

TL;DR

This work addresses the need for continuous, privacy-preserving traffic monitoring on bridges by combining vision-generated labels with SHM sensor networks. It introduces a two-stage pipeline where CV outputs label SHM data for training a graph attention network that performs multi-task regression to estimate vehicle counts by type and lane, enabling camera-free monitoring after deployment. On a real Swiss bridge, the method achieves up to 99% accuracy for light vehicles and 94% for heavy vehicles, demonstrating robust performance across acceleration and strain inputs and reducing reliance on permanent visual instrumentation. The approach repurposes existing SHM infrastructure for traffic analysis, offering a scalable, privacy-friendly alternative with strong industrial feasibility and potential applicability to various bridge geometries.

Abstract

Bridges, as critical components of civil infrastructure, are increasingly affected by deterioration, making reliable traffic monitoring essential for assessing their remaining service life. Among operational loads, traffic load plays a pivotal role, and recent advances in deep learning - particularly in computer vision (CV) - have enabled progress toward continuous, automated monitoring. However, CV-based approaches suffer from limitations, including privacy concerns and sensitivity to lighting conditions, while traditional non-vision-based methods often lack flexibility in deployment and validation. To bridge this gap, we propose a fully automated deep-learning pipeline for continuous traffic monitoring using structural health monitoring (SHM) sensor networks. Our approach integrates CV-assisted high-resolution dataset generation with supervised training and inference, leveraging graph neural networks (GNNs) to capture the spatial structure and interdependence of sensor data. By transferring knowledge from CV outputs to SHM sensors, the proposed framework enables sensor networks to achieve comparable accuracy of vision-based systems, with minimal human intervention. Applied to accelerometer and strain gauge data in a real-world case study, the model achieves state-of-the-art performance, with classification accuracies of 99% for light vehicles and 94% for heavy vehicles.

Figures (13)

• Figure 1: Problem definition
• Figure 2: At least four pairs of points on the bridge ($X_w, Y_w$) and corresponding pixels in the video frame ($p_x, p_y$) are used to solve the approximate projection between the plane of interest and video frames. The region of interest within the bridge is assumed to be flat.
• Figure 3: With continuous vehicle counts, each vehicle is proportionally assigned across consecutive signal segments
• Figure 4: The architecture of the CNN encoder
• Figure 5: Definition of: True positive: A; False positive: B, C; False negative: D. The larger the overlap area $A$ relative to the total area, the better the alignment between the predicted and true values.