Three-dimensional Reconstruction And Disaster Identification Of Slope Using Unmanned Aerial Vehicle-based Oblique Photography Technique

Slope disaster is one of the main diseases in engineering construction,which is characterized by strong suddenly and great difficulty in prevention and control.It is easy to cause traffic interruption,river blockage,building collapse and other serious safety accidents,thus leading to heavy property losses and casualties.The traditional manual inspection method of slope has the problems of high risk,difficulty and low efficiency,and it is difficult to serve the long period and high frequency slope disaster investigation.In this study,artificial intelligence technologies such as machine vision,photogrammetry and unmanned aerial vehicle(UAV)system were used to carry out intelligent slope disaster monitoring method,and a 3D reconstruction and disaster identification method based on UAV oblique photography was proposed.This slope monitoring method will improve the inspection efficiency,reduce risk,reduce the difficulty,have important engineering practical value and social value.The main research contents are as follows:(1)Research on 3D reconstruction method of slope model based on Structure from Motion with multi-view Stereo(Sf M-MVS)algorithm.Firstly,based on the slope photography,the aerial photography parameters such as the height,image overlap degree and slope Angle were set to obtain the multi-view image of the slope.Combined with the scale-invariant feature transform(SIFT)algorithm,the beam method aerial triangulation algorithm,and the PMVS algorithm,the 3D point cloud model with the point spacing less than 1mm was reconstructed.Finally,by constructing irregular triangular meshwork,encapsulating 3D white film and texture mapping,the 3D real scene model with an overall accuracy better than 3.0cm and an accuracy of millimeters in X and Y directions was reconstructed.The three-dimensional point cloud model constructed based on the SFM-MVS method in this paper can achieve the same density as the three-dimensional laser scanning technology.The accuracy of 3D real scene model is much higher than the industry standard.(2)Study on slope hazard identification method based on Multiscale Model to Model Cloud Comparison(M3C2).Firstly,on the basis of point cloud sampling,Statistical Outlier Removal(SOR)algorithm was used to remove Outlier points from point cloud to obtain high-quality basic point cloud.Furthermore,combining with the engineering experience method and the Iterative Closest Point Algorithm(ICP)Algorithm,the two-stage point cloud with position error up to several meters is successfully registered to centimeter level.Finally,the M3C2 point cloud analysis algorithm was improved to optimize the radius parameters of normal vector fitting,to automatically identify the variation in the slope area,and to achieve the quantification of the variation value.(3)Research on engineering application of 3D modeling and disaster identification method based on UAV oblique photography.In a slope project in Hunan Province,first of all,images before and after slope disasters were obtained based on tilt photography.Then,the millimeter-scale 3D point cloud model was reconstructed to build a 3D real scene model with an accuracy better than 2.7cm.Secondly,the highquality registration point cloud was obtained by preprocessing,and the M3C2 algorithm was introduced to successfully identify the slope disaster area and its variation.Finally,cracks with a height difference of 1.38 cm,a distance of 2.87 cm in the potential danger area were extracted by combining with the 3D real scene model.In a slope project in Jiangxi Province,the density millimeter 3D point cloud model was reconstructed,and the accuracy of the 3D real scene model was better than 2.0cm.Secondly,according to the slope engineering design drawings,the point cloud data of multiple sections before the slope disaster is constructed.Furthermore,the M3C2 algorithm was used to compare and analyze the point cloud data of section before and after the disaster,and the position of maximum deviation and irregular deformation area were determined.Finally,combining with the 3D real scene model,it is found that there are fractures with a width of about 11.0cm in the irregular deformation area.