Long-range vision-based displacement sensing for structural health monitoring of fire-exposed spatial structures

Abstract Structural health monitoring (SHM) during fire requires deformation measurements that remain reliable when contact sensors and cabling cannot be deployed safely. This study presents a long-range visionbased system for measuring nodal displacement in fire-exposed suspendome structures. The system couples an enhanced YOLOv11 detector with StrongSORT tracking and an adaptive displacement (AD) reconstruction algorithm based on inverse perspective transformation and template matching. Two multiscale attention block configurations target complementary imaging regimes: small targets against complex backgrounds and low-contrast targets obscured by smoke. Validation used five targets on a scaled suspendome and co-located cable displacement sensors during medium-and high-power fire tests at a 100 m stand-off distance. Under medium-power exposure, the system resolved displacement variations of approximately 1 mm. Under high-power exposure, the maximum mean absolute error was 1.322 mm, and peak-displacement error remained below 5% for all nodes. Compared with displacement computed directly from bounding-box centers, AD reduced peak error, mean-square error, and weighted mean absolute percentage error at every target. The results establish an experimentally validated remote sensing layer for SHM under fire. They also show that spatially non-uniform smoke is the dominant limitation and should be represented by channel-level visibility and confidence metrics in field deployment.