Fusion Of Aerial Imagery And Airborne LiDAR Data For Post-earthquake Building Point Cloud Extraction

Devastating earthquake is one of the most miserable natural disasters to affect mankind.The building is the most important hazard-affected body in post-earthquake disaster.Building destruction is the main source of caused casualties and economic losses.Traditional information extraction of earthquake damages for buildings is based on artificial ground sampling survey which lasts longer and needs a lot of human resources.With the rapid development of remote sensing technology,especially the launch of high-resolution remote sensing satellites for civilian use and the rapid development of low-altitude unmanned aerial vehicle(UAV)technology since the beginning of the 21 st century,the high spatial resolution optical satellite images are widely used to get ground two-dimensional information and the generated three-dimensional information of stereo pairs can't meet accuracy requirement.However,the earthquakes cause the collapse of buildings and change some information such as building height.Airborne laser radar system can obtain highly precise three-dimensional elevation information.Airborne laser radar data is applied to extract the information of post-earthquake building by using elevation texture information which is derived from two-dimensional raster data generated by the point clouds interpolation.And the existing research shows that some of the original characteristics of point cloud are conducive to identify earthquake-induced building damage.But these literatures did not introduce how to extract building point clouds in the seismic field.Extraction of post-earthquake building point clouds is difficult to distinguish collapsed buildings and vegetation because they have similar characteristics of point cloud.The spectral information could be introduced to distinguish buildings and vegetation because they are quite distinct.Aerial images and airborne LiDAR(light detection and ranging)points cloud are obtained at the same time.They have the same location data and the geometric registration is easy to be implemented.Therefore,in order to identify single building damage by using characteristics of point cloud,this thesis carried out the method research that extraction of post-earthquake building points cloud based on fusion of aerial imagery and airborne LiDAR data.In order to extract post-event building point cloud,this thesis mainly studied two methods and key technologies which include the DBSCAN(Density-based Spatial Clustering of Applications with Noise)Clustering algorithm and object-oriented method.The main research contents include the following.(1)This thesis studied that applicability and robustness of frequently-used point cloud filter methods in the earthquake-stricken area.We took post-event airborne LiDAR points cloud in the Haiti 7.0 magnitude earthquake as experimental data.The experiment have chosen four filtering methods,such as mathematical morphology filtering,slope filtering,TIN(Triangular Irregular Network)filtering and cloth simulation filtering(CSF),to conduct filtering experiment in the quake-hit regions.The results show that mathematical morphology filtering and CSF filtering are better of filtering effect.Their Kappa coefficients are 0.86 and 0.85,respectively.(2)The spectral information of aerial image was assigned to points cloud to distinguish between building and vegetation points.The spectral characteristics of building and vegetation points were analyzed.This article proposed a new method to determine threshold segmentation,minimum hybrid probability method.The results show that this method has better segmentation effect than the traditional bimodal threshold segmentation method.At the same time,the experiment analyzed the extraction of vegetation by VDVI(visible-band difference vegetation Index),GRDI(green-red difference index)and single band.And we find that red band has better separability than any other single band.(3)There are still a large number of discrete noise after extraction of vegetation by point cloud segmentation method.For the first time,the DBSCAN(Density-based Spatial Clustering of Applications with Noise)clustering algorithm,based on the spatial density,was used to extract post-earthquake building points.The results indicates that this method is feasible.There are two important parameter(neighborhood radius ? and minimum points)in the clustering method.We determine two important parameter by a semi-automatic method.And when the neighborhood radius ? is 4 meters and minimum points threshold is 20,the sum of?error and?error is minimized.(4)For further use of image information and points cloud space feature information,this thesis introduced object-oriented method to extract building points cloud.Firstly,optimal segmentation scale is applied to multi-scale segmentation.Then,according to the grey value features of red band,geometric and elevation features,building points cloud are obtained using the rule-based classification,adjacent-supervised classification and fuzzy classification method.Finally,the extracted building points cloud are compared with referential point cloud data.And Kappa coefficient of the three methods are 0.84,0.70 and 0.84,respectively.Therefore,it shows that object-oriented rule classification or fuzzy classification method to extract building point cloud is a kind of good method in the earthquake disaster area.In addition,the data of test area was experimented.The results show that rule-based classification and fuzzy classification method possess a good robustness.