Research On Building Recognition Based On Structural Feature And Stereoscopic Feature

Buildings are one of the most important artifacts and are closely related to people's lives.In the data mining,unmanned,precise guidance and many other areas for the identification of buildings have a potential demand.The traditional build i ng recognition technology have poor performance when in the face of low-resolut io n images,complex structure or similar shape of the building by using two-dimensio na l data,so the use of three-dimensional data to identify the building has impor ta nt research value and far-reaching significance.In this paper,we mainly study the stereoscopic recognition method of buildings,which are divided into four types: ridge shape,chevron shape,flat top and cone shape according to the top type of common buildings.By using DSM data with resolution of 0.1m,we extract the structural characteristics and the stereoscopic characteristics of the object for stereoscopic recognition.In this thesis,the three-dimensional structure of four types of roof buildings is analyzed.In view of the phenomenon that the DSM data contains a large number of singular points and noise points,an improved regional growth algorithm is used to segment the top and the initial topology is obtained by edge extraction.On the basis of this,the morphological method is used to fill the initial topology,and then the deformed contours are geometrically corrected by Douglas-Pock method.Finally,the topological structures are rotated calibrated and scaled respectively.Aiming at the key problems in target recognition,that is,the rotation,scale and translation invariance of the feature,the paper studies a point pair histogram featur e.Firstly,the normal vector of the target surface is estimated by K-dimensional tree,Knearest search and principal component analysis.According to the high noise characteristics of large-scale buildings,the stereoscopic characteristics suitable for small target recognition are improved,and the stereoscopic recognition is carried out under the condition of different scale of target and multi-angle target samples.The recognition results of different characteristics in different noise environments verify the resolution and robustness of the characteristics of different types of buildi ngs.For the individual recognition of the established target,the paper studies a 3-D shape context feature,and improves the feature by using the spherical harmo nic transformation to solve the problem of rotatio n invariance.Finally,the feature matching is carried out under the same sample condition in different characteris t ic parameters,the optimal parameters and recognition precision are obtained,and the rotation invariance of the feature is also verified.At the end of this thesis,the corner and the contour of the of the topologi ca l structure are mergered,then generate the eigenvector.The principal compone nt analysis method is used to reduce the dimensionality of point pair histogram featur e.Through the synergistic of structural features and point pair histogram features,the types of buildings are identified in different noisy environments.The recognit io n result verifies the anti-noise capability of the synergistic feature,which has highe r accuracy and robustness than using the stereoscopic feature only to recognize.