Search

In the present structure health monitoring work, regular monitoring of the deformation and displacement of the target building structure is an important means to evaluate the safety of the structure. Aiming at the difficulties of traditional measurement methods such as complex operation, high cost and too few monitoring points, this paper proposes a method based on computer vision to obtain 3D coordinates of surface feature points of building structures. Based on the principle of restoring the target structure from the camera motion, the sparse point cloud model of the target building is initially reconstructed and the attitude information of each camera is obtained simultaneously. The depth information of each pixel in the original image is obtained by trial matching based on the principle of pole line search, and corrected and optimized based on the consistency of optics and geometry. Based on the original sparse point cloud, the depth information of each image was fused and reprojected based on the depth map registration principle to reconstruct the dense point cloud model of the target building. Finally, the 3D coordinates of the target building surface points are obtained and measured based on the scale factor method. In order to verify the feasibility of the method, seven buildings of different sizes were selected for field measurement and compared with manual measurement results. The results show that the measurement results of the proposed method are almost the same as those of manual measurement, and the error level fully meets the requirement of the error limit of the general structural health monitoring work. The method described in this paper obtains a large number of 3D coordinates of detection points on the surface of building structures in a non-contact form at one time, improves the flexibility and intensity of measurement point layout in the actual measurement work, and its operation is simple and low cost. • 3D measurements of buildings were made by multi-view shots from a single camera. • Combined with image depth information, the density of point cloud model is improved. • The accuracy of coordinates is improved by optical and geometric constraints. • Obtain a large number of coordinates at a time, making point selection more flexible. • Low cost, simple non - contact measurement, error to meet the accuracy requirements. [ABSTRACT FROM AUTHOR]