Abstract: To address the challenges of high cost and difficult deployment of traditional structural displacement sensors in complex environments,and to validate the applicability of vision-based measurement methods in practical engineering,a real-time displacement monitoring system based on monocular vision sensing and deep learning technology was developed.The system uses a low-cost CMOS camera to perceive a specially designed target.An intelligent recognition module based on the YOLOv11 algorithm enables real-time target detection.Ellipse fitting technology is employed to extract subpixel coordinates of the target center,and actual displacement is calculated through scale conversion.To mitigate measurement anomalies caused by on-site wind disturbances,the median absolute deviation (MAD)algorithm is introduced for error compensation.Compared with single-BDS displacement monitoring results,the system achieved a root mean square error (RMSE)of less than 1 mm in both horizontal and vertical directions,significantly outperforming the single-BDS method (RMSE of 1.88 and 1.57 mm in the plane direction,and 3.28 mm in the elevation direction).At monitoring distances of 20,50,and 75 m,the MAD compensation algorithm improved measurement accuracy by 26.6%,24.2%,and 35.6%,respectively.Experimental results demonstrate the excellent performance of the proposed vision-based sensing system in complex environments.It provides a low-cost,high-precision,non-contact technical solution for structural health monitoring of slopes,buildings,and other engineering structures,showing high potential for engineering application and promotion.

Key words: vision measurement, displacement monitoring, YOLOv11, error suppression, engineering application, surveying and mapping technology