Smart structural health monitoring using multi-camera computer vision and edge computing for long-span bridge mode shape identification

Accurate displacement measurement is critical for reliable structural health monitoring (SHM) and modal identification of large-scale structures. Computer vision–based displacement measurement provides the advantages of long-distance monitoring, non-contact sensing, and easy deployment, making it well suited for civil infrastructure SHM. Nevertheless, its large-scale use remains constrained by the substantial data volume generated from high-frame-rate video and the difficulty of achieving precise synchronization among multiple cameras, which hinder continuous multi-point monitoring on large structures. This study develops and validates an advanced multi-camera edge-computing vision-based 2D displacement measurement system (EdgeCVDMS-II) for synchronized, real-time multi-point displacement tracking and mode shape identification. The system integrates six high-resolution IMX477 cameras with adjustable zoom lens and an NVIDIA Jetson AGX Orin edge device, enabling low-latency GPU-accelerated template matching with subpixel refinement. Laboratory validation on a 4 m long simply-supported bridge model demonstrates the displacement measurement accuracy of 0.15 pixel, with a normalized root-mean-square error (NRMSE) of 1.45% relative to laser displacement sensors and excellent agreement with accelerometer-based modal parameters. Field deployment on a 280 m long span cable-stayed bridge confirms the system’s capability to capture spatially dense low-frequency dynamic displacement responses under ambient excitation at over 100 m stand-off distance. The identified fundamental mode shape achieved strong correlation with accelerometer-derived results (MAC = 0.9724), demonstrating robustness low-order modal identification and serviceability-level structural assessment under real-world conditions. The proposed EdgeCVDMS-II overcomes key limitations of existing vision-based methods by providing synchronized, distributed, and real-time video processing and displacement measurements, extending the applicability of computer vision–based SHM to continuous monitoring of large-scale civil infrastructure.