Abstract: Monocular visual settlement monitoring represents a novel non-contact monitoring approach. Nevertheless, challenges arise from intricate backgrounds and stringent benchmark target installation prerequisites, occasionally impeding the efficacy of this method for settlement monitoring. This paper introduces a non-contact monocular vision method for high-precision automatic settlement monitoring. Initially, the measurement target is positioned within the designated monitoring area, and target identification is achieved utilizing the YOLOv10s target detection algorithm. Subsequently, the edge-based least squares ellipse fitting technique is employed to determine the pixel coordinates of the target's center point. Following this, leveraging the principles of camera imaging, a derivation process is conducted to establish the world coordinates corresponding to the pixel coordinates of the target's center point using an enhanced world coordinate computation approach. Ultimately, the actual displacement value is computed with the initial frame image of the monitored area serving as the reference. Through outdoor simulation settlement experiments, we compare the monocular visual calculated value with the electronic level measurement value, using their absolute error as the evaluation metric. By varying the distance between the camera and the target, we assess the monitoring accuracy of the proposed method. At a distance of 5 m, the maximum absolute error is 2.784 mm, and the minimum is 0.246 mm. When the distance is increased to 10 m, the maximum absolute error rose to 4.071 mm, with a minimum of 0.42 mm. The experimental results demonstrate that as the distance between the target and the camera increases, the accuracy decreases. However, within a distance range of up to 10 m and a settlement value of 250 mm, the average absolute error of the method stands at 1.543 mm.

Key words: monocular vision, settlement monitoring, YOLOv10s, pixel coordinates, world coordinates, electronic level