Research On Methodology Of Lidar Data Filtering And Building Extraction

LiDAR is a new technique in acquiring ground information, becoming a researcharea bearing significant application value in remote sensing. Based on LiDAR pointcloud data, this paper illustrates the working system and data obtaining process ofLiDAR, further studies point cloud outlier removal, filtering and finally uses supportvector machine as the basic theory for point cloud building detection.The work of the thesis can be divided into the following parts:The first part mainly demonstrates the system components, ranging and locatingprinciple as well as the data type point cloud. Then some current popular LiDARsystems are listed and compared. After analyzing the sources and distribution of LiDARpoint cloud outliers, the PCA3D shape analysis is applied to detect and eliminate thesefoot points.On the basis of the first process, adaptive TIN filtering is then implemented,resulting in two sets, namely ground points and non-ground points. This method hassimple and clear mathematic principles. More importantly, it can cope with variousground and target types. Especially because of the first step, filtering can achieve betterresults.Finally, a new idea of building detection based on support vector machine isproposed. Since filtered point cloud only contains building, vegetation and other objects,the paper uses the rest point cloud as inputs of classification. Appropriate features arefirstly selected, including height texture acquired from Gabor filters. This practicesolves the discrimination between buildings and high vegetation. The next step is usingsupport vector machine as the classifier, chooses samples for training and testing andsegments building areas. Experiments demonstrated the feasibility and value of thisbuilding detection tragedy.This paper only deals with LiDAR point cloud without other data resources,combining advantages of various data processing and image processing algorithms, andaccomplishes the job of building detection, which has some positive influence onfurther research of LiDAR data processing.