Computer Vision Method For Monitoring Deformation Of Ancient Buildings

Chinese ancient architecture is a consolidated memory of 5,000 years of historical development and social revolution in China,reflecting the comprehensive attainment of the Chinese nation in architectural skills,culture and art.Throughout their long historical service cycle,ancient buildings have suffered different levels and types of damage.The overall and localised deformation of ancient buildings is the most critical and serious of all types of damage,which seriously threatens the structural health of Chinese ancient buildings.There is an urgent need to adopt advanced monitoring techniques for long-term deformation monitoring of ancient building structures.This paper explores the application and development of the latest computer vision techniques in artificial intelligence methods for monitoring the structural health of ancient buildings.In response to the deformation monitoring needs of ancient buildings,a vision system applicable to the overall deformation monitoring of ancient buildings,the development of growth deformation monitoring techniques applicable to cracks in Tibetan ancient mural walls,and the construction of an intelligent mural wall crack semantic segmentation model are developed,and the main research content of the paper is as follows.(1)For the overall deformation monitoring needs of ancient building structures,a deformation monitoring technique for ancient building structures based on computer vision methods was studied.Firstly,the limitations of existing techniques in ancient building monitoring applications are improved by introducing the Zhang Zhengyou camera calibration method and sub-pixel interpolation.Secondly two types of motion tracking theories,template matching and feature point matching,are used to monitor the deformation of the structure.Finally,the effectiveness of the technique in planar vibration and static monitoring tests of ancient building models is investigated,and the effects of different tracking theories,monitoring distances and monitoring angles on the visual monitoring system are analysed.(2)For the long-term deformation monitoring needs of the wall cracks of Tibetan ancient buildings,the long-term growth and deformation monitoring technology of the cracks of Tibetan ancient building murals based on computer vision method is studied.Firstly,the images of cracks from different perspectives obtained from camera intervals are corrected by image alignment algorithms.Secondly,a variety of image pre-processing operations are applied to remove the long-term monitoring interference,and the crack features are extracted using the image processing threshold segmentation method,and feature information such as crack contour area,contour perimeter,density,centre of gravity,crack skeleton line length,average width and maximum inner joint circle width are calculated.Finally,the growth of cracks was measured by multiple indicators based on the information of various crack characteristics.(3)In view of the limitations of crack extraction by threshold segmentation in longterm crack growth and deformation monitoring,a semantic segmentation model for cracks in ancient Tibetan architecture frescoes based on deep learning methods is studied.Firstly,the network structure of U-Net model is improved to construct a semantic segmentation model for cracks in the wall of Tibetan ancient building murals.Secondly,the crack semantic segmentation model is used in the long-term growth and deformation monitoring of cracks in ancient building murals.Finally,the performance of the two crack deformation monitoring methods in the long-term monitoring of cracks in Tibetan ancient building murals is compared and analysed.There are 102 figures,13 tables and 122 references in this dissertation.