Geometric Structural Matching And 3D Reconstruction Of Buildings

The automation of building scene photogrammetry,recognition and positioning is based on the application of robots and machine vision.Feature matching is a key and difficult step in the process of machine vision algorithm implementation.The selection of features,the accuracy and repetition of feature matching affect the accuracy of target recognition,the accuracy of image mosaic,the calculation of camera attitude and the effect of 3D reconstruction.The feature description and 3D information reconstruction of buildings play an important role in the automation of information collection and spatial layout planning in urban space.In view of the structural characteristics of man-made scenes,through the analysis of geometric features,line features on building structures are used as matching objects to extract structural information.Linelet algorithm is used to extract the building frame,and then the natural scenery line segment is removed to improve the proportion of building features.Based on the characteristics that the line feature of building is at the intersection of two walls,a line segment matching method is proposed,which combines the texture feature of the line segment neighborhood with the structural feature.The method has the properties of rotation invariance,scale invariance and projective invariance.The method of line group matching based on the texture inside the angle overcomes the texture similarity problem caused by building symmetry,and can effectively meet the transformation situation in building information acquisition.The matched line features are reconstructed to obtain the 3D information of the building frame.Taking line feature data as object,the camera attitude matrix is solved by using trifocal tensor relation.Then,the error between the reconstructed line segment and the measured value in Hough space is used as the objective function of bundle adjustment to optimize the model.3D reconstruction based on line features can effectively improve the computational efficiency,and can obtain higher building structure information integrity in fewer acquisition images.