Research On Construction Methods Of High-precision Surface Deformation Field Based On Monocular Vision

Deformation monitoring is a significant part of engineering and industrial survey.Traditional deformation monitoring methods often need to contact the measured objects,accompanied by the shortcomings of low efficiency and high labor cost.Besides,by using traditional methods,we can only obtain single-point monitoring results,which cannot reflect the the spatial state and the deformation trend of the deformable bodies comprehensively.Since monocular-vision monitoring system was proposed,it has gradually become a research hotspot in the field of deformation monitoring in recent years due to its non-contact,low cost,high efficiency and easy operation.Based on the monocular close-range photogrammetry technology,the high-resolution deformation image sequences acquired by the system and pre-processed by effective relevant algorithms are used to construct the high-precision deformation field for the surface of the deformable bodies.The area-wise spatial characteristics in the deformation process of the monitored objects are described comprehensively,which could provide new ideas for the deformation monitoring methods.The main works of this paper are as follows:(1)The composition and measuring principle of monocular-vision deformation monitoring system are introduced,and the main coordinate systems and their transformation in monocular vision system are elaborated;the principle of Zhang Zhengyou's camera calibration technology is introduced detailedly,and the image distortion correction formulas are deduced so as to eliminate the radial distortion of monocular images effectively;by optimizing the quality of images with noise and missing partial pixel,including image denoising,image enhancement and sub-pixel interpolation,the high-quality data sources for image deformation extraction can be provided in the later stage.(2)Three main image displacement extraction methods are systematically elaborated: digital image correlation algorithm based on the gray value(DIC),frequency-domain phase only correlation algorithm(POC)and SIFT/SURF feature point extraction and matching algorithm.In particular,POC can be used to estimate the overall rigid body translation of the image;using DIC method can obtain deformation data in the form of regular dense grids;the fused SIFT and SURF feature points can be extracted and matched to obtain scattered high-precision displacementvectors covering the deformed area of the images.A weighted function of correlation coefficient considering the RGB channels is proposed to improve the DIC algorithm,which has improved the accuracy and reliability of DIC,and the superiority of the method is verified by experiments.(3)Effective processing of outliers in deformation field is realized by threshold filtering and median filtering;the deformation field with incomplete data is reconstructed by bicubic B-spline interpolation algorithm;the surface deformation field is visualized by using layered pseudo-color image and vector field image,which shows the deformation state and trend of the deformable bodies intuitively.(4)Through the experiment of monocular-vision deformation monitoring,the monitoring of a simulated landslide surface was carried out.After image pre-processing,the two-dimensional surface deformation field was constructed by DIC algorithm and SIFT+SURF feature point matching algorithm respectively,and compared with the real displacements of check points measured by high-precision total station.The experimental results showed that the accuracy of the deformation field constructed in this paper can reach sub-millimeter level within the monitoring distance of 10 meters approximately and under the specified hardware configuration,which verified the feasibility and reliability of the proposed methods for high-precision visual deformation monitoring.