Digital construction demands scalable structural health monitoring that is accurate, low cost, and deployable across infrastructure.
Researchers at Chonnam National University in South Korea have introduced a virtual sensor grid method that advances vision-based structural health monitoring for infrastructure and construction assets. Led by Professor Gyuhae Park from the Department of Mechanical Engineering, the research replaces pixel-level sensing with superpixel-based virtual sensors, improving robustness, interpretability, and reliability while maintaining a low-cost monitoring approach suitable for digital construction environments.
Structural health monitoring (SHM) underpins the safety and reliability of assets across civil engineering, aerospace, and industrial sectors. In construction and infrastructure monitoring, vibration-based SHM is widely used to detect damage by identifying changes in structural vibration characteristics. Conventional approaches rely on contact sensors such as accelerometers, which present constraints around installation, coverage, cost, and spatial resolution. These limitations restrict scalability across large or complex assets, particularly in live infrastructure environments.
Non-contact alternative
Vision-based structural health monitoring has emerged as a non-contact alternative, enabling full-field vibration measurement directly from video data. By removing the need for surface-mounted sensors or physical markers, vision-based methods reduce deployment costs and support monitoring of structures with limited accessibility or complex geometries. For digital construction applications, this aligns closely with the drive towards data-rich, sensor-light monitoring strategies.
However, existing vision-based SHM methods often rely on pixel-level motion estimation. While technically effective, pixel-based approaches are sensitive to noise, lighting variation, low-texture surfaces, and image distortion. Even phase-based optical flow methods, which estimate motion from phase information rather than intensity changes, remain vulnerable due to their reliance on individual pixels. This can limit their reliability for infrastructure monitoring in uncontrolled site environments.
Challenges
To address these challenges, Professor Park’s research team developed a superpixel-based virtual sensor framework that aggregates pixels with similar structural and vibrational behaviour. “Our approach utilizes superpixels, clusters of neighboring pixels with similar vibrational and structural behavior, as virtual sensors for motion estimation,” explains Prof. Park. “This creates a virtual sensor grid that can adapt to any structure and offers robust and accurate full-field vibration measurement without the need for physical markers or contact sensors.”
The framework operates through a three-stage process. First, pixel-level motion is extracted from video sequences using the phase nonlinearity-weighted optical flow (PNOF) algorithm, previously developed by the authors. This method evaluates the reliability of displacement estimates in different directions, discarding components affected by high phase nonlinearity. The result is a marker-free, full-field displacement map based on reliable motion data.
Confidence value
In the second stage, the method calculates an overall confidence value for the displacement estimated at each pixel. This provides a built-in reliability assessment, distinguishing this approach from other vision-based vibration measurement techniques that lack confidence metrics. For infrastructure monitoring and digital twin integration, this reliability layer is essential for trustworthy decision-making.
The third stage groups pixels into superpixels using both displacement information and confidence values, forming a virtual sensor grid aligned with the monitored structure. Depth information is incorporated to improve spatial alignment between the sensor grid and the physical asset. Displacement data is then calculated at the superpixel level, enabling damage detection using virtual sensors rather than individual pixels.
Experimental validation
Experimental validation on an air compressor system demonstrated that the superpixel-based virtual sensors achieved accuracy comparable to a laser Doppler vibrometer. While individual pixels showed variability, aggregating them into superpixels reduced noise sensitivity and improved measurement stability. Importantly, this was achieved without physical markers or contact sensors, reinforcing the suitability of the approach for low-cost infrastructure monitoring.
“Vibration-guided superpixel segmentation enhances both robustness and interpretability of structural diagnostics even in complex environments,” says Prof. Park. “Our approach makes full-field structural monitoring accessible, low-cost, and deployable using ordinary cameras supporting applications in infrastructure monitoring, aerospace and mechanical equipment diagnostics, smart cities, robotics, and digital twins.”
Practical step
Published online on 30 September 2025 and appearing in Volume 240 of Mechanical Systems and Signal Processing on 1 November 2025, the study positions superpixel-based virtual sensors as a practical step forward for vision-based SHM. For the digital construction sector, the method supports scalable monitoring strategies aligned with asset management platforms, digital twins, and smart city infrastructure.
By replacing pixels with superpixel virtual sensors, this framework strengthens vision-based monitoring, supporting digital construction workflows, safer infrastructure decisions, and broader adoption of camera-based condition assessment methods.
Read next: Digital twins as risk visibility platforms across the asset lifecycle – Build in Digital
Are you a building professional? Sign up for a FREE MEMBERSHIP to upload news stories, post job vacancies, and connect with colleagues on our secure social feed.