|其他摘要||Point cloud is a collection of mass points that express target spatial distribution and surface properties in the same space coordination. Point cloud generated by density matching of oblique aerial images is similar with point cloud from LiDAR(Light detection and Ranging, LiDAR), both of which include three-dimensional coordinates information of features. Compared with the latter, the former doesn’t contain the information of multi-echoes and echo intensity, but its cost is much lower than the latter. It is a hot issue in current research how to gain interesting features’ three-dimension model, for instance buildings, from point cloud. Due to noise and tree blocks in point cloud from oblique aerial images, three-dimension coming from them have problems such as the blur edges and contradictory expression of boundaries and so on. Besides most of digital productions of aerial photography currently stop at DSM,4D production and the three-dimension model buildings, there is little further analysis. Based on this, how to optimize point cloud from oblique aerial images to generate models with obvious edges and logical boundaries is studied in this thesis, further research was done to estimate floor area ratio of different regions in study area using three-dimension information of building roofs.
The research object is point cloud from oblique aerial images around Turpan city of which the area is about 150 kilometer, against where there are much noise and uninteresting features like trees and cars, this paper analyze the difference and contact between segmentation and classification of point cloud, several methods to extract buildings and main ways of data filters, study how to product building digital surface model of cities, which get rid of blur edges and disturbing from other features, making use of characteristics of point cloud from oblique aerial images. In addition, based on the three-dimension information of building roofs, floor area ratio of different regions in city is estimated. The main study works are as follows:
(1) Extraction of three-dimension information of building roofs. Three-dimension information of building roofs includes height and area information. Potential building areas are gained through slope value by calculating aspect image and slope image of nDSM(normalized Digital Surface Model, nDSM), affiliating constrains of height threshold, texture information and shape threshold. Roof planes are gained in advantage of least-squares fitting plane. Finally K-means is used to optimize building boundaries, producing accurate building boundaries. Compared to the result of several traditional edge detectors, the latter is disturbed by all kinds of other features whose edges are detected. Through combination showcase of point cloud and fitting roof planes, quantitative comparison combining corner height, method in this paper has higher horizontal and elevation accuracy in three-dimension information of building.
(2) Produce of BDSM(Building Digital Surface Model, BDSM).Unconstrained D-triangulation of DEM(Digital Elevation Model, DEM) is generated by method of point-to-point interpolation. Considering the extracted building boundaries as the constrained edges, embedding them into unconstrained D-triangulation, we get constrained D-triangulation. Triangulations only covering building areas gained through cutting out DSM-triangulation by edges of buildings. In the end BDSM with obvious edges can be got by combining triangulation of building with unconstrained triangulation of DEM.
(3) Calculation of floor area ratio of city. Floor area ratio of different regions in study area is calculated with three-dimension information and area of different districts. The experimental result, which is in line with the distribution regulation, compared to shadow –based method, has a higher accuracy through field-proven.
This paper focuses on the BDSM through optimizing of point cloud and extraction of city’s floor area ration, which are of importance to further build of accurate building model. The work done in this paper offers a more rapid way to display and update data for urban construction and planning department, which has a certain meaning both theoretically and practically.|