Research On Building Extraction Algorithm Based On High Resolution Aerial Image And LiDAR Point Cloud

Building is one of the most important ground features in the urban area.The accurate extraction of building is of great significance for urban planning,3D modeling,population estimation and making and updating topographic map.Therefore,it is a popular research direction in the field of remote sensing image processing.Meanwhile,how to extract all the buildings more accurately,more efficiently and more automatically is also a difficulty and challenge in the research.With the continuous improvement of UAV(Unmanned Aerial Vehicle)photogrammetry and LiDAR(Light Detection and Ranging)technology,it provides powerful technical support for the acquisition of more accurate 3D coordinates and other kinds of geographic information,and provides plenty of available data for building extraction.Based on high resolution aerial image and LiDAR point cloud data,this paper combines the rich spectral features of high resolution aerial image with the accurate elevation features of LiDAR point cloud.By combining the advantages of the two kinds of data,the automatic urban building extraction is realized according to the special features of buildings,which are different from the other ground objects.The main research work in this paper is as follows:(1)The raw point cloud is preprocessed and the DSM grid is generated.The raw point cloud contains a lot of noise,so the method of statistical filtering is used to remove the high noise points and the low noise points.According to the size of the relevant high resolution image,the grid of the same size is generated,and then the spatial interpolation of the point cloud is carried out to obtain the DSM grid.And the DSM is smoothed as the basic data for subsequent processing.Since the DSM is the same size as the image,the two kinds of data can be fused according to the grid coordinates.(2)The spectral features of vegetation and buildings are extracted base on the high resolution aerial image.Through the analysis of vegetation spectral features,VI(Vegetation Index)is formed for the IR-R-G three-channel image and R-G-B image.By analyzing the characteristics of each VI in the different research data,the threshold segmentation based on the Otsu algorithm is carried out to extract the vegetation areas.By evaluating the accuracy of each vegetation areas extracted by different VI method,the optimal VI is selected to make vegetation mask to eliminate the negative effect of vegetation.Based on the spectral features of buildings,two building indices,MBI(Morphological Building Index)and EMBI(Enhanced Morphological Building Index),are used to extract buildings.By comparing the extraction results of the two indices,the extraction result by EMBI is more accurate.However,the completeness of the result is affected by the shadow interference.So mark the centroid points of the building patches extracted by EMBI and more favorable information can be provided for subsequent building extraction.(3)The buildings are extracted based on probabilistic voting and morphological region growing.According to the definite edge character of building,Canny algorithm is used to locate the edge in DSM grid.For each edge point,adopt certain strategy to obtain effective spatial voting points based on the elevation characteristics of the building edge.The kernel density estimation(KDE)is used to carry out probabilistic voting statistics for each voting point and mark the local maximum points in the entire probabilistic value.The local maximum points,along with the centroid points of the building patches extracted by EMBI,are used as seed points.The morphological region growing is realized by thickening the boundary of seed region iteratively.An appropriate height difference threshold is set according to the elevation features of the roof planes,which is used to judge whether the growing segment is object area.Combine all the object areas obtained from the growth of each seed point and refine the result to get the final building areas.(4)Based on the above research work,MATLAB R2014 a is used as a tool to realize the proposed algorithm.And five groups of experiments from Vaihingen region and Xianning region are conducted to verify the performance of the proposed algorithm.By analyzing the accuracy of building extraction results and comparing with different algorithms,the feasibility and reliability of the proposed algorithm are verified,as well as the high completeness,high correctness and high quality especially when extracting buildings with a large area.