Extraction Study Of Urban Buildings Based On Airborne LIDAR Data

Today, the density of population in the urban area is very high and there are many production activities in the city. Building model can effectively reflect the city's space. So it may be useful to choose an effective method to extract building models. Airborne LIDAR technology is a new measurement technique, it can collect the LIDAR point cloud data from the surface quickly and accurately. So it can extract building models of the urban areas. Compared to using frame imagery to extract the building models, using LIDAR technology has many advantages:Firstly, LIDAR is an active sensor, it could work without other light sources and has less affected by the external environment, so it can work all day. And optical measurement methods due to greater impact on the external environment and can not work all day. Secondly, LIDAR system can obtain the densely LIDAR point cloud data(5-10 points per square meter), so the data it obtained is accuracy and the edge characteristic line is stable. The edge characteristic line of the optical image will be interferenced by light and shadow, so it could show the characteristics of instability. Thirdly, for those which using optical measurement methods are difficult to distinguish such as dense forest areas, dense urban areas and desert areas, using LIDAR technology could distinguish them better. In general, use airborne LIDAR technology to extract the building model is an accurate and efficient method.However, the lacking of automatic building model extraction algorithm greatly baffled the application of airborne LIDAR technology. In this thesis, after deeply investigate the key technology for automatic building model extraction from airborne LIDAR data, a new method for automatic extraction buildings is proposed. The major works includes:1. Describe the knowledge of LIDAR technology, including its working principl, applicable scope and characteristics of LIDAR point cloud data. Then, it introduces several typical filtering algorithms and building extraction methods of LIDAR point cloud data.2. A rapid classification algorithm based on strip partition is proposed. Because of building points belong to off-terrain points, the first thing to do is to divided the LIDAR point cloud data into ground points and off-terrain points. The implemented method divides the points into small stripes in x and y direction, as there are polynomials in x and y direction every point is classified twice. Only the points that are classified as off-terrain candidates twice are considered to be off-terrain, others are ground points. Points that are classified different in x and y direction often can be found at steep slopes.The method this paper is proposed reducing the dimension through the stripe partition, so the speed of classification is increased. Experiment results show that, compared to progressive TIN filtering algorithms and slope filtering algorithm this method has a lower total error rate and the results is more accurate.3. A method of building segmatation and edge detection based on delaunay triangulation is proposed. After the points have been classified, points that belong to one building and points belong to building edge have to be identifid. Find points belong to one building is done by a segmentation. Most of the intrinsic segmentation algorithm need to selected seed point, but it is difficult to choose a suitable seed point. Most of the intrinsic LIDAR data edge detection algorithm use detetion algorithm based on image edge. But image edge detection algorithm only suitable for GRID, which use on LIDAR data the edge it detected different from the real edge. In view of intrinsic building segmentation algorithm and edge detection algorithm have deficiencies, segmentation and edge extraction of irregular LIDAR point cloud data need to seek other options.In this thesis, a new method of segmatation and edge extraction based on LIDAR point cloud data is proposed. The LIDAR cloud points data are connected by a Delaunay-Triangulation. The triangulation establishes the neighbourhood relation between the points. Then, all triangles have been assigned to a segment. Finally, after removing trees by the area control, then small group of noise points be deleted.Using culture area of changchun 4 square kilometers LIDAR point cloud data as empirical data. Experiment results show that using the method this paper is proposed could detect most building points from the LIDAR point cloud data. And this method could detect all building edge and the Boundary line is Smooth. So the method can meet the needs of building extraction.