Research On Refinement Extraction Of Urban Buildings Combining LIDAR Point Clouds And Orthographic Images

With the rapid development of modern remote sensing technology,the means of obtaining remote sensing information has become more and more abundant,and the quality and accuracy of data have been greatly improved.As one of the most important feature categories in urban areas,building information has important research significance in urban construction planning,road planning,real estate survey,and digital city modeling.However,based on spectral remote sensing image data,it is desirable to obtain The building category of three-dimensional information still has great difficulties.The rich three-dimensional coordinate data obtained by LiDAR measurement technology will make up for the lack of information of planar image data.In this paper,a digital surface model is created after processing LiDAR point cloud data,and the contour extraction of buildings is studied and analyzed by combining DSM with high-level information and high-resolution orthophotos.The main research contents and results include:(1)Preprocessing the noise point of the laser point cloud data to prepare for the rasterization of the discrete data.The spatial grid is established and a variety of interpolation methods are compared.The interpolation method with high precision and good effect is searched to fit the digital surface model,and the median filtering process is performed on the produced model to improve the accuracy of the model.(2)The process of extracting the main contour of the building is: using the Slope(Zevenbergen,Thorne)slope extraction algorithm to obtain the slope information of the model according to the high-precision digital surface model,and using the contrast splitting algorithm in the slope information.The slope change region is segmented,and then the body object of the building is extracted based on the feature change of the adjacent height difference between the objects.(3)Combine spectral images with object-oriented ideas to refine the boundaries of buildings.In the traditional multi-scale segmentation based on spectral bands,the boundary recognition model extracted by Canny edge detection is integrated to optimize the segmentation object.The spectral difference features are created and the neighborhood features are combined to analyze the outline of the building.Finally,the accuracy of the urban vegetation that is incorrectly divided into building objects is removed.(4)The final result extracted is compared with the building object delineated by manual visual interpretation for accuracy analysis and evaluation.The accuracy evaluation indexes used are correct rate: 90.14%,integrity: 96.72%,quality: 87.74 %,the overall accuracy of the building information extracted according to the above steps is ideal.