Research On Key Technology Of Building Extraction With DOM Assisted By Airborne LiDAR

City is the center of economic、cultural、information and political of a regional, and also it is radiation and accumulation of human civilization;so informationize trend is the core of the global informationize. As an important part of the city, building plays an important role on building the city, such as urban planning, mobile communications network, disaster incident management, noise and air pollution analysis, and the protection of cultural heritage etc. Especially in today’s whit rapid development of urban construction, city building is changing. So fast, one of the most important things in map updating and building three-dimensional reconstruction is urban buildings extraction rapidly.Airborne LiDAR as a new type of Earth observation technology have an advatage of accessing three-dimensional surface features information directly compared with the traditional remote sensing images. LiDAR point clouds with high vertical accuracy (15-30cm), highly automated with data processing, have low level accuracy, with the lack of texture and structure information, so using it alonly to extract buildings has highly automated, but more errors. The aerial ortho-photograph with rich texture and structure information, has high level accuracy (up to centimeter-level), and can extract buildings with high precision, but it is difficult to process automately and manual intervention is still required.The main research objective is to extract the original LiDAR point cloud building as the prior knowledge of DOM assisted extraction of high resolution structures, in order to achieve automatic and accurate extraction of building purposes. The main contents of the research and innovation of this paper are:(1) A summary of the research situation on the theory of building extraction at both home and abroad is given; the key problems that need more research on building extraction are pointed out.(2) Based on the analysis of several commonly used edge-detecing operators in remote sensing image, this paper deeply researches and analyzes the Gabor wavelet, which has been difficultly applied in processing of large remote sensing imagery confining to its own processing speed. To solve this problem, we propose two fast edge detecting methods based on Gabor wavelet. One is the parallel fast Gabor wavelet edge detecting method based on FFTW, whose idea is to transform the convolution of Gabor wavelet to product operation of frequency domain, and at the same time using FFTW (West Fast Fourier Transform) in the frequency domain processing; the other is based on the anti-symmetry of the imaginary part of Gabor wavelet. It proposes a simplified Gabor wavelets method, which quantifies different levels of Gabor wavelet to generate simple symmetrical templates, and which uses 8 threads to do parallel computing Gabor wavelets simplified in two frequencies and four directions. So the speed of edge detection is greatly improved;(3) This paper provides a method of rapid DEM generating based on Mean-Shift, compared with the commonly used method of DEM extraction based on mathematical morphology and surface fitting. Based on above analysis, the normalize digital surface model (nDSM) is generated, and then buildings are detected by comprehensive utilization of elevation,variance,density of point clouds,first and last echoes and so on.(4) Considering that the building points detected in LiDAR pointcloud is irregular and random, this paper provides a method of rapid DEM generating based on Mean-Shift, compared with the commonly used method of DEM extraction based on mathematical morphology and surface fitting on the basis of studying the Douglas-Peuker algorithm and the tube algorithm.(5) The paper presents a method to extract the outline of buildings from the DOM by using the information of position and direction of outline of bulidings from LiDAR pointcloud. In this method, three kinds of information are used to constraint the standard Hough transform to quickly detect candidate segments and get their direction, distance and other information in polar coordinates. Then by using the smallest pixel edge comiectivity solutions, it can quickly generate the candidate segments set in DOM.(6) For the candidate segments from the buildings generated in DOM, an outline-filter method based on the density analysis of LiDAR point clouds is well used to extract the building outline accrately.