Monitoring And Analysis Of Artificial Rainfall-type Landslides Based On Trinocular Vision Technology

In the landslide monitoring research,the appearance of deformation characteristics usually means the slope becomes instable,so it is important to obtain continuous landslide deformation data for landslide analysis.The traditional way to get data is from remote sensing images and aerial data mainly considers the multi landslide in the large-scale spatial range.The cost of data acquisition is high,and the interval between remote sensing and aerial imagery is long,which cannot provide continuous slope deformation data.And the data is lack of detail for a single slope.In this paper,the indoor physical rainfall model test is designed to simulate the rainfall conditions under natural environment,and simulate the landslide of different morphological slopes under different rainfall conditions.At the same time,a trinocular camera monitoring system based on binocular vision was introduced to obtain 3d depth information of the slope.Based on the system,an image recognition method based on superpixel theory for landslide deformation region is proposed.This method can automatic extracted landslide area,obtain continuous deformation data,and combine the depth information to provide 3D surface reconstruction.The specific research results are as follows:(1)Based on the traditional indoor rainfall landslide test,a new monitoring method was designed.The trinocular camera was used to collect the depth information of the slope.At the same time,the pore water pressure sensor and camera equipment were combined to monitoring the slope deformation.With 2D and 3D data obtain by the monitoring system,a wealth of slope deformation information data was obtained.(2)Using Python and OpenCV open source library to automatic extract the landslide region based on superpixel theory.The initial segmentation of images by superpixel and the change of slope image features during landslide deformation define a color based feature.The image roughness is defined,and achieved an automatic threshold selection method.The method can automatically extract the landslide deformation region information from the image,so as to accurately identify the characteristics of the landslide monitoring image and its shape spatial change information,and analyze its changing trend.The depth information collected by the trinocular camera restores the slope point cloud data and performs 3D reconstruction,and the accuracy is verified by comparing with the actual slope model size.By comparing the difference and influence of 2d landslide data and 3d space attributes on landslide monitoring analysis,it is proved that the 3d information can describe and express the slope instability process more accurately,and obtain more abundant landslide deformation information.(3)The 2D area change trend is obtained by the image recognition result,and the volume change of the slope deformation is calculated in the 3D space according to the 3D surface reconstruction result.At the same time,the slope failure process is analyzed by the pore water pressure change data.The effects of different rainfall intensities and different slope morphologies on slope stability were compared.The differences and effects of 2D landslide deformation information and 3D spatial attributes on landslide monitoring analysis were compared.The results prove that 3d information can be more accurately described and expressed.The slope instability process.(4)Analyze the error part of the recognition result,start from the position of the error and the image features,then comparing the actual physical test results,and determine the camera resolution,the change of the specular reflection and the background mutation are the main causes of the recognition error.