Font Size: a A A

City Building Detection From Multiple Aerial Images

Posted on:2012-04-01Degree:MasterType:Thesis
Country:ChinaCandidate:Y LinFull Text:PDF
GTID:2218330362453612Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
Digitally replace the traditional manual measurement is the current trends in the field of surveying and mapping. Compared to other digital methods, aerial images benefit for its large amount, information-rich and low-cost access. Reconstructing three-dimensional information from aerial images is a good method to obtain urban geographic information. We studied building detection methods for high-resolution aerial images in urban area, and the problem of deformation, blocking and interference derived. Base on that, we propose and build a system for detecting rectangle buildings in urban area, the system features the following:1. Improved the belief propagation method for stereo matching. In our approach we use the energy decrements between the gray differences of two pixel window instead of the gray difference of two pixels. This reduces image noise and the effects of discontinuity. Our approach also limits the propagation using image segmentation method, which improves the matching speed and accuracy.2. We use a dense stereo matching point cloud to replace the traditional DSM. And simulate the Priming Model of human recognition process, using the hypothesis-validation method to detect buildings. In the hypothesis phase, we use the matching point cloud to compute Gaussian curvature of the ground area, and then combine the elevation of ground points and Gaussian curvature to filter ground point as hypothesis seed points. And then a region growing method is run to establish the initial building hypothesis. In the validation phase, we use Hough transform to extract line from hypothesis, use cluster analysis to group lines to two main directions as the validation of building hypothesis.Feature works include the optimization of stereo matching algorithms. Belief propagation can be optimized by the GPU architecture or CPU + GPU hybrid architecture. And the texture of a building have features in frequencies, we can use it to verify a building. In addition, the shadow of building, vertical lines can also be added to verifying conditions.
Keywords/Search Tags:stereo vision, image matching, belief propagation, building detection, three-dimensional reconstruction
PDF Full Text Request
Related items