| This dissertation is concerned with building extraction from UAV high resolution image data. Buildings in UAV high resolution image data have the characteristics of a large area in the building, edge feature increasingly long, rich fine details on the rooftop of building . In order to extract building in UAV high resolution image data, we respectively improve the existing edge detection algorithm and line detection algorithm. Finally, we utilize the relationship between lines to build line structure graph to extract building.The main works and innovations are listed as follows:(1) Buildings in UAV high resolution image data have the characteristic of rich fine details on the rooftop of building. In this paper, we improve canny edge detection, improved canny edge detection can automatically select the scale, with the purpose of suppressing rich fine details on the rooftop of building and highlighting the main edge feature image.(2) Traditional Radon transform which transform the whole image into polar coordinates,in this way, traditional Radon transform has good performance in the image which contains few significant lines. but for UAV high resolution image data this method doesn't have good performance. In this paper, we improve Radon transform searching window, which makes the algorithm more robust. Traditional Radon transform can not provide endpoint information, we utilize the least square fitting to obtain the endpoint information line.(3) Combined with knowledge of buildings, line structure graph is given to extract buildings from line subset. First, we initially connection line subset. Second, to improve the recognition system of the computing speed. In this paper, we use cluster analysis algorithm cluster line set in Radon space. Third, build line structure graph to extract closed path. Finally, filter closed path with building structural features. Experimental results show that this method can effectively extract building from UAV high resolution image data. |
PDF Full Text Request