Bridges rarely fail without warning. Long before a serious structural problem becomes visible to the public, traffic, weather, moisture, and repeated cycles of loading can leave behind cracks, concrete spalling, and water leakage. Detecting how these defects change over time is therefore essential for protecting infrastructure and preventing costly emergencies. Yet the most common inspection methods remain labor-intensive, expensive, and sometimes dangerous, requiring engineers to work near traffic, at heights, or above water. A new artificial intelligence-powered system developed by researchers in South Korea could make this process faster, safer, and far more precise by turning routine drone images into a long-term record of bridge damage.
The research team, led by Assistant Professor Hyunjun Kim of Seoul National University of Science and Technology, created an automated computer vision framework that can compare images collected during separate bridge inspections, even when the drone was not in exactly the same position. This is a major challenge in infrastructure monitoring. Photographs taken weeks or months apart often differ in camera angle, distance, lighting, and viewpoint. As a result, software may struggle to determine whether it is seeing the same crack or damaged region, whether a defect has expanded, or whether a newly detected mark is simply an imaging artifact. The new system is designed to overcome these inconsistencies by anchoring every later inspection to a common three-dimensional reference model.
The process begins during an initial inspection, when a drone captures multiple overlapping images of the bridge. These images are used to reconstruct the structure in three dimensions through photogrammetric techniques, which estimate the position and shape of surfaces by analyzing how they appear from different viewpoints. Instead of rebuilding a new 3D model every time the bridge is inspected, the researchers use this first model as a persistent spatial reference. Images gathered during later flights are then matched to locations on the original model, allowing engineers to revisit the same structural components and compare their condition over time.
To align the new photographs with the reference model, the framework combines hierarchical localization and image clustering. Hierarchical localization narrows down where a photograph was taken by first identifying the relevant section of the bridge and then refining the camera’s position and orientation within that area. Image clustering groups photographs that show similar portions of the structure, helping the system distinguish between different bridge elements and viewpoints. Together, these techniques allow the artificial intelligence to recognize corresponding regions despite changes in the drone’s flight path or viewing angle. The result is a more reliable comparison between inspections that would otherwise be difficult to align manually.
The system also incorporates Global Navigation Satellite System data to convert measurements made in images into real-world dimensions. This is important because a crack or spalled area can appear larger or smaller depending on the camera’s distance and perspective. By combining geographic positioning information with the 3D reconstruction, the framework can estimate the physical size of a damaged region rather than merely reporting how many pixels it occupies. Engineers can therefore track changes in the length, width, or surface area of defects using measurements that are meaningful for maintenance decisions.
The researchers tested the method on an in-service prestressed concrete bridge over a 120-day period using drone imagery collected during routine inspections. The bridge provided a realistic test environment in which the camera viewpoint changed between surveys, reflecting the conditions that engineers commonly face in the field. During the study, the framework tracked cracks, areas of concrete spalling, and water leakage as they developed. It was able to identify and follow the progression of damage even when the images were captured from different distances and angles, demonstrating that the reference-model approach can support monitoring across multiple inspection dates.
Its measurement performance was particularly notable. When the system’s estimates of damaged areas were compared with conventional manual measurements, the maximum error was 4.61 percent. That level of accuracy suggests that drone-based imaging could provide useful quantitative information without requiring inspectors to physically access every affected surface. More importantly, the system does not simply detect whether damage is present. It creates a visual and measurable history of how the damage changes, helping engineers distinguish stable defects from areas that may require urgent intervention.
The framework could also reduce the computational and operational burden associated with long-term monitoring. Conventional approaches may analyze each inspection independently or require a new 3D model to be generated whenever fresh images are collected. By relying on one reference model, the new method can maintain consistency from one survey to the next while avoiding repeated reconstruction. This could make it easier for transportation agencies to establish regular monitoring programs, particularly as aging bridges require more frequent assessment and maintenance budgets remain limited. A consistent digital record could also support predictive maintenance, allowing repairs to be scheduled before deterioration becomes severe.
The researchers acknowledge that the method is currently best suited to relatively flat bridge components. Highly curved surfaces may introduce additional geometric and visual challenges, potentially reducing measurement accuracy. Even so, the approach can cover many of the bridge elements commonly examined during routine inspections. The team believes the same principles could eventually be adapted to tunnels, dams, elevated rail systems, and other infrastructure where repeated visual surveys are needed. By combining drones, 3D reconstruction, satellite positioning, and artificial intelligence, the technology offers a glimpse of infrastructure management in which bridges are not merely inspected at isolated moments, but continuously observed through a growing digital history of their condition.
Subject of Research: Bridge damage monitoring using drone-based computer vision and 3D reconstruction
Article Title: Long-term monitoring of damage progression using multi-view images from routine inspections of bridges
News Publication Date: 27-Apr-2026
Web References: https://doi.org/10.1177/14759217261443618
References: Structural Health Monitoring, DOI: 10.1177/14759217261443618
Image Credits: Assistant Professor Hyunjun Kim, Seoul National University of Science and Technology
Keywords
Bridge monitoring, artificial intelligence, drone inspection, computer vision, structural health monitoring, 3D reconstruction, infrastructure safety, crack detection, concrete spalling, predictive maintenance
Tags: AI-based infrastructure inspectionautomated damage detection in bridgesbridge damage monitoringcomputer vision for civil engineeringcost-effective bridge inspection methodsdamage progression analysisdrone image analysis for bridgeslong-term bridge maintenance technologylong-term structural health monitoringnon-invasive bridge safety assessmentsafe and efficient bridge assessment toolsSouth Korea AI infrastructure research
