Indoor 3D Reconstruction Based On Global Matrix And Constraint Optimization

In recent years,with the development of 3D laser measurement technology,3D point cloud has been widely used in indoor scenes,such as indoor mapping,indoor navigation and positioning,3D reconstruction,interior decoration and many other fields.3D laser measurement technology is not only rapid data acquisition,but also high precision,can accurately obtain the details of indoor space information,the maximum restoration of the status quo of indoor space.However,3D point clouds have disadvantages such as large amount of data,unstructured and lack of semantic information.Therefore,more and more scholars are studying the reconstruction of 3D models with complete structure and rich semantics based on 3D point clouds.However,compared with the outdoor environment,the indoor environment contains a large number of furniture and sundries that block the laser measurement,resulting in the serious absence of indoor three-dimensional point cloud.In addition,for indoor Windows,glass walls and other high-transmission structures,laser scanning error is large,and point cloud noise is serious.Therefore,there are still many problems to be solved in the existing methods of reconstruction of indoor 3D point cloud structure model: 1)Due to the absence and noise of indoor 3D point cloud,the existing methods are prone to over-segmentation,under-segmentation or missegmentation when conducting indoor 3D point cloud plane structure segmentation;2)Existing methods of plane structure feature fitting are all aimed at single plane fitting,and lack of fitting constraints between planes(parallel or vertical constraint relations).The structural constraints of the final reconstruction model are weak,while the interior structure all follows certain rules and constraints.3)The existing application scenarios of interior structure model reconstruction methods mostly follow the rules of Manhattan World(MW)and Weak Manhattan World,and there are few researches on the reconstruction of non-Manhattan interior scenes.To solve the above problems,a method based on global matrix and constraint optimization was proposed to reconstruct 3D point clouds in indoor scenes with nonvertical inclined plane structure.Firstly,aiming at the plane structure extraction of indoor three-dimensional point cloud,an improved RANSAC point cloud plane structure segmentation method is proposed.The method adopts the idea of RANSAC model fitting to initially segment indoor point clouds to obtain local units,and introduces weight items in the fitting process to realize the segmentation of local units at different scales.Then,the normal vector,bounding box and other features of the local unit are calculated.According to the idea of Region Growing characteristic growth,the features of the local unit are judged and the connectivity analysis is added to achieve the segmentation of the plane structure from "part to whole".On the basis of plane structure extraction,a plane regularization method based on global constraint matrix is proposed.The method introduces the constraint relationship between indoor structures(vertical or parallel constraints)in the process of plane fitting,converts the least square problem of plane fitting into matrix form,and transforms the problem of minimizing error into the problem of matrix optimization decomposition.Then,the global constraint matrix was constructed according to the vertical parallel constraint relation between planes,and the corresponding parameters of each plane were solved by optimizing the global constraint matrix to achieve plane regularization.Finally,on the basis of plane regularization,a topology reconstruction method based on constraint optimization is proposed.Firstly,the occupancy probability raster diagram of indoor point cloud is generated,and the connected space is extracted for raster optimization.Then,semantic labels are added to the raster map with morphological method to achieve the effect of room segmentation,and then the raster room and 3D point cloud are superimposed to achieve the monomer of point cloud room.Finally will each room of the plane to expand intersecting candidate set of surface structure,and according to the candidate surface fitting,features of candidate surface intersection and the effect of neighborhood points of candidate supporting function of the structure about the candidate can choose the constraints of the optimization function model,the topology reconstruction problem is transformed into function optimization problem,reconstruction of topological relations between the flat structure,Finally,the reconstruction work from indoor threedimensional point cloud to structural model is realized.This study implements from complex scenes of indoor structure model of 3 d point clouds reconstruction,and in a number of real data sets to experimental verification method,the results show that the research method to rebuild the model structure of the complete geometry,topology relationship closely,strong binding rules,apply to Manhattan,weak Manhattan and the non-Manhattan indoor scene(not a vertical slope structure of indoor scene).