The Key Technologies For The Extraction Of Buildings From Oblique Aerial Images

Buildings are important objects in urban areas,and 3D(three dimensional,3D)building models play an important role in land mapping,urban planning,disaster management,emergency response and other applications.With the development of"smart city",the demand for 3D building models-especially the 3D building models which have the semantic information-are increased day by day.The scholars at home and abroad have done a lot of research on how to extract buildings.Both the Lidar point cloud and high-resolution images are the important data sources for building extraction.However,the Lidar point cloud is not easy to access.The cost of the Lidar point cloud is much higher than that of the high-resolution aerial or satellite images;so the high-resolution images are still the most important data sources of building extraction.Compared with the vertical aerial image and satellite image,the oblique aerial images have the characteristics of visible and less occlusion of building facade,and is a high quality data source for building extraction.Therefore,it has great research value and social benefit to study how to extract buildings by the oblique aerial images.In fact,from the perspective of the photogrammetric processing,the DIM point cloud,DSM and orthophoto data can be generated from original aerial images,and all these data can provide various features for building extraction.So,the fusion of the features from DIM point clouds and the features from images for the extraction of buildings becomes a worthy research topic.However,compared with the Lidar point cloud,the noise of the DIM point cloud is larger,and the differences of the shape features of buildings and tree are smaller.All these drawbacks damages buildings extraction by the fusion of the features from DIM point cloud and the features from oblique aerial image.To solve this problem,we focuses on the study on how to extract the buildings using the oblique aerial images.This paper mainly contain four parts.?First,this paper summarizes the difference between DIM Point cloud and Lidar Point Cloud,and focuses on the distribution of DIM point cloud and the applicability of the algorithm.The experimental results show that the distances from the point to the fitted plane is distributed as the normal distribution in the local area of the DIM point cloud.At the same time,we select the same geometric features and applied classification method based on random forest to classify the DIM point cloud and the Lidar point cloud.respectively.The experimental results show that this algorithm which is applied for the Lidar point cloud performs poorly when the DIM point cloud is classified.It is verified that the difference between the DIM point Cloud and the Lidar point cloud.?DIM point cloud filtering is an essential step of building extraction.In order to solve the problem that the performance of progressive triangular irregular network(TIN)densification(PTD)filtering method which ignore the fact that both the standard variance and density of DIM point clouds are large and the DIM point clouds can't penetrated into the plaits and touch onto the ground is poor,this paper proposed an improved PTD filtering algorithm for DIM point clouds.First,a new strategy of seed points selection based on the facades of the building is adopted by this method aimed at the fact that the DIM point clouds cannot penetrate into the vegetation and touch to the ground in the stage of seed point selection.Then,this method makes use of the iterative densification strategy of the PTD algorithm to densify the(TIN)constructed by the seed points.When the density of the points belonging to TIN is greater than the given threshold,a new iterative densification scheme based on multi-scale is used to continue to densify the TIN.The difference between the improved method and PTD method is the change of densification angle threshold which is from small to large during the second densification phase.The experimental results show that the improved PTD algorithm can effectively separate the ground and non-ground points respectively.The performance of our improved method for DIM point clouds is better than that of the PTD algorithm;hence,this method can be one of important tools for DIM point clouds.?In the stage of building detection,we present a framework for detecting and regularizing the boundary of individual buildings using a feature-level-fusion strategy based on features from dense image matching(DIM)point clouds,orthophoto and original aerial images.The proposed framework is divided into three stages.In the first stage,the features from the original aerial image and DIM points are fused to detect buildings and obtain the so-called blob of an individual building.Then,a feature-level fusion strategy is applied to match the straight-line segments from original aerial images so that the matched straight-line segment can be used in the later stage.Finally,a new footprint generation algorithm is proposed to generate the building footprint by combining the matched straight-line segments and the boundary of the blob of the individual building.The performance of our framework is evaluated on a vertical aerial image dataset(Vaihingen)and two oblique aerial image datasets(Potsdam and Lunen).The experimental results reveal 89%to 96%per-area completeness with accuracy above almost 93%.Relative to six existing methods,our proposed method not only is more robust but also can obtain a similar performance to the methods based on LiDAR and images.?In the stage of building modeling,the semantic information of building roof is extracted by combining DIM point cloud and images,and the automatic semantic modeling process only using the oblique aerial image is discussed based on the existed semantic modeling framework.In this paper,a buildings extraction method only based on the oblique aerial images is proposed and this method obtained good results.This work is of great significance to extract the buildings using the oblique tilted aerial images.