Research On Appliances Of Bayesian Networks In Based-Geometry Detection Of Building

It's an important research to extract 3-D buildings from 2-D intensity images in Computer Vision. It is also a complicated question. At present many studies depend on pure computer graphics, which little integrate with the knowledge of artificial intelligence and less make use of the reasoning of bayesian networks. In this thesis we adopt the reasoning of bayesian networks to detection of building.Artificial intelligence technique can accelerate this research process to obtain better efficiency. As is known to all, 2-D image lacks direct 3-D information.This paper describes how to recognize objects in simple 2-D image with bayesian networks. We first complete a bayesian networks inference algorithm based on message-passing, then present an improved line extract method (ILEM)based on image basal feature, which makes localness integrate with wholeness. And we discuss how to get the unknown camera parameters from a single aerial image.This algorithm can deal with image low-level information by the inference of bayesian networks. We make 2-D hypotheses from the intensity line features extracted from the input image by an edge detector. We can last get the parameters of 3-D model by selecting and verifying the features in 2-D image with bayesian networks. The experiment shows that the technique is feasible and efficient.