| The building damage caused by earthquakes is the major factor of casualties and economic loss,and the building damage degree investigation is the important basis of disaster loss assessment.Many cases of remote sensing applications in earthquake emergency proved that satellite and airborne remote sensing with high resolution have already become significant methods for disaster monitoring.Light Detection And Ranging(LiDAR)is a new active remote sensing technology,which can quickly obtain ground elevation data with high precision,and can be used to monitor secondary disasters such as surface deformation,landslides and dammed lakes,and damage of roads and buildings.The main research methods of building damage detection using LiDAR data are: 1.Change detection based on the height difference of point cloud before and after the earthquake or the generated Digital Surface Model(DSM)height difference image.2.Extraction from the fusional images of post-earthquake point cloud or DSM and highresolution optical images.3.Building damage analysis by three-dimensional spatial parameters of post-earthquake point cloud such as intensity,echo times,slope,aspect and local height difference.The first method requires the accumulation of preearthquake LiDAR data in stricken areas,which makes it impractical for the actual earthquake emergency and disaster monitoring work.Meanwhile,the problem about high accuracy registration of the multi-resource image fusion exists in the second kind of methods.The third kind of method is the important direction for future research,however the earthquake damage detection parameters with high precision are relatively insufficient.Aiming at the current challenges of building seismic damage based on airborne LiDAR,this work,from the view of practicality,studies and develops effective quantitative detection characteristic parameters of building damage,without the assistance of pre-earthquake data,and merely using post-earthquake airborne LiDAR point cloud.This study is expected to promote the quantitative identification technology of building earthquake damage based on LiDAR data,and improve the accuracy of earthquake disaster assessment.In order to study the effective characteristics of building damage extraction and damage degree quantitative detection on the basis of high accuracy airborne LiDAR data after earthquakes,this thesis carries out the studies of the airborne LiDAR point cloud filtering method,building point cloud extraction method and the single building point cloud coordinate correction method.The standards of remote sensing buildings classification and interpretation signs of images and point cloud are established.A new method: extract roof point cloud section by section along with the longitudinal axis(length)and analyze the earthquake damage sensitivity of three-dimensional spatial characteristics and transverse profile(width)similarity of each section,is proposed.The characteristic parameters of sectional roof damage identification and the model of building damage degree detection are created.The characteristics and models established in this work are tested by the sample point cloud of the airborne LiDAR after the Mw 7.0 earthquake in Haiti on January 12,2010.The main research contents and the advancement are as follows:(1)A method for extracting building point cloud with the fusion of airborne LiDAR and high-resolution imagesThe prerequisite of carrying out the research of building seismic damage is to separate the building point cloud from the whole LiDAR data.On the basis of investigating the current situation of Laser point cloud filtering technologies,the typical building damage areas of Port-au-Prince in the Haiti Mw 7.0 earthquake on January 12,2010 are chosen as study areas,and Cloth Simulation Filter(CSF)is selected as the filtering method.Based on airborne LiDAR point cloud,the ground points extraction is studied.After several experiments,grid size of 1.2m,classification threshold of 0.50,roughness of 0.3 and iteration number of 200 is the best parameters setting for the CSF method in this research areas,which reduces the probability of the debris of the collapsed buildings being wrongly classified into ground points and reduces the computational complexity relatively.The technical flow of object-oriented building point could classification with the fusion of high-resolution optical images and airborne LiDAR data is set up,by using the mutation characteristic that the collapsed buildings after earthquake no long have regular edge height and the phenomenon of multiple echoes appearing in the house boundary.Based on the two characteristics,Normalized Vegetation Index(NDVI)and height,the normalized Digital Surface Model(nDSM)is divided and classified in a large-scale to remove roads,bare land,vegetation and small surface features.The large-scale segmentation objects are further segmented and classified to remove roads,bare land and vegetation with smaller size,and finally obtain the building point cloud.Compared with visual interpretation based on 0.15 m aerial images,the results of automatic classification show this method is of high precision.(2)A new coordinates automatic correction approach of the single building point cloud is presented.The damage characteristics of buildings studied in this paper are basis of the equal interval sampling roof point cloud of the single building according.Therefore,establishing a point cloud coordinate system of the single building provides a solid foundation for the normalization and automation of characteristic analysis.Actually,the LiDAR could points are mostly represented in the geodetic coordinates,where the X-axis is not completely consistent with the house strikes(the longitudinal direction).Thus,we propose a new method of coordinates automatic correction of single building point cloud.The originof corrected point cloud coordinate system locates at a certain corner of single building.At the same time,the X-axis,Y-axis and Z-axis are distributed along the longitudinal direction(length),horizontal direction(width)and vertical direction(height)respectively.The key to coordinate correction is the algorithm,which can automatically determine the outline of the building and the angle between the longitudinal with horizontal directions.First of all,all building point clouds are projected to the horizontal plane.What's more,a convex polygon is determined using the Delaunay triangulation(DT subdivision)method.Nevertheless,under the influence of sample selection and distribution of point cloud,there is a minor error between the direction of long axis of automatic determined convex polygon and the actual bearing of the building(the angle between the vertical and the X-axis).To eliminate the error,according to the characteristic that the horizontal outline of most buildings is rectangular,the minimum area algorithm is employed to determine the boundary rectangle of the convex polygon,which is the closest to the shape of the building.Next,the longitudinal direction of the building is automatically determined via the length and width of the rectangle.Finally,the new coordinates of single building point cloud are obtained after the coordinate system rotation.(3)Interpretation keys to building damage of remote sensing images and airborne LiDAR point cloud are established.Combining with buildings classification of and grading of building damage in the ground survey,according to recognition capability of remote sensing images and airborne LiDAR point cloud,individual buildings are classified as six categories of high-rise buildings,multi-stores buildings,general houses,large-open buildings and other type.And the seismic hazard degrees of single building could be divided into collapsed(?),partly collapsed(?)and not collapsed(?).When the recognition capability of remote sensing images is good enough or with the assistance of elevation data,the third degree(?)can be further divided into obvious damage(?a)and no obvious damage(?b level).On this basis,remote sensing image interpretation signs,which include different buildings types and damage degree,are established in accordance with the anomaly characteristics of building geometry,spatial layout,spectrum and shadow.Further,the airborne LiDAR point cloud characteristics of different damage degree are established based on height changes of the whole building and the profile,spatial distribution of point cloud,intensity change and echo times.The established interpretation keys of different buildings types and levels of building damage,which involve remote sensing images and airborne LiDAR point cloud,and typical images will provide important reference standard for earthquake damage investigation and damage assessment of buildings.(4)The damage feature parameters and a method of detecting seismic damage based on the similarity of projection points of buildings' sectional roof section are proposedIn order to find more effective characteristic parameters of buildings' earthquake damage recognition,we use the similar characteristics of the projection of sections' roof of the undamaged buildings,and put forward a detect method of building damage by a similarity neighboring comparison of lines connecting the projection points in the section.The error of point cloud measurement may lead to the increase of vertexes,which will affect the similarity of graphics' size and shape.Therefore,Douglas-Peuker(D-P algorithm)simplified algorithm is employed to cut down the vertexes number of the profile graphic.Given that the 0.15 m precision of the test data of Haiti earthquake airborne LiDAR,a conclusion is obtained that when the distance threshold of the D-P algorithm is 0.3m,the simplified figure is closest to the projection shape of the building section in many experimentsOn the basis of comparing and analyzing the features of the vertex number,area,perimeter and centroid position distribution of simplified polygons,which represent the section plane of various kinds of buildings including not collapsed,partially collapsed and collapsed,similarity matching models of simplified graphics are established.The models include the shape similarity model represented by the number of vertexes or the compaction,the size similarity model represented by area or perimeter and the location similarity model represented by centroid distance.It is found that the number of vertexes and the similarity of perimeter are better than the compaction and area similarity in tolerance and applicability,in addition,the shape,size and position similarity can all reflect the damage of the house.The multi-feature minimum space distance approach of the integrated similarity features of shape,size and position is used to identify whether the subsection roofs are damaged or not.According to the damage degree of each section,the damage level of building is quantitatively assessed.Moreover,using the samples from Haiti earthquake airborne LiDAR test data,the total accuracy of the similarity assessment method is about 89% comparing to the visual interoperated damage level.(5)Three-dimensional characteristic parameters and seismic damage detection method based on the segmented building roof are proposed.The existing detection methods of building damaged degree based on Lidar data mainly use the spatial features such as distance,area,angle,volume,slope,and aspect etc.,which are calculated from the points within a neighborhood,and according to the correlation coefficients of these features before and after the earthquake to decide the buildings damaged degree.However,the accuracy of point clouds registration and data availability limit the practicality of the change detection method.This paper proposes three-dimensional characteristic parameters of segmented building roof based on postearthquake point clouds,and depend on which establishes the extraction method and seismic damage quantitative assessment method.The buildings are divided into several roof sections at equal intervals in the longitudinal direction of the building.The 3D point cloud characteristics of each section of the roof are extracted.Utilizing the characteristics that the roof slope of the building is basically the same,we analyze the relationship between the spatial characters changing and the damage degree.Based on this research idea,the characteristic parameters and calculation model of building segmental roof,such as surface area,volume,and extension coefficient ? are established.Based on the theoretical relationship between volume and surface area,the difference d? between the extension coefficient of the surface area estimation and the extension factor of the volume estimation was proposed as a feature to detect the roof damage.The variation of the theoretical extension coefficient d? at different slope gradients with 2%,10%,20%,and 30% changing is estimated.According to the theoretical extension coefficient d?,the examples of buildings with flat roof and slope roof,the threshold values to divide the undamaged roof,obvious damage,severe damage,and complete destruction is determined.Based on the damage of each roof section,use the established roof damage ratio parameters to quantitatively assess the damage ratio of the building.According to this,46 individual building samples are selected for experimental analysis.The test results show that the overall detection accuracy can reach 91% compared with the visual interpretation result,and the difference between the extensional coefficients can detect the part that is not collapsed but has obvious damage.In addition,since the density of the point cloud will affect the three-dimensional spatial characteristics,the interpolation method based on triangulation is mainly studied based on the analysis of the existing point cloud interpolation method.We analyze the effect of interpolation methods such as nearest neighbor method,linear method and cubic convolution on the surface area,extension coefficient,volume and roof height of building segmented roofs.Comparing with different methods,we get the conclusion that the linear method is a better point cloud interpolation sampling method,which not only maintains the same distribution pattern of features calculation results between the interpolated features and the original point cloud,but also keeps the similarity of each building segments,and highlight the significant differences of damaged part. |
PDF Full Text Request