Indoor Space Decomposition And Automatic Method For Indoor Modeling From Laser Point Clouds

The latest development of laser measurement technology has inspired substantial interest in indoor modeling.More and more applications require three-dimensional modeling of indoor scenes,including indoor navigation and positioning,emergency services,crowd management,building planning and simulation.Building structure modeling using indoor three-dimensional point clouds is an important means to construct city information model.Indoor space has some characteristics different from outdoor space,such as multi-storey and closure.Because of the constraints of interior building structure,interior space is often expressed as cellular space,showing the characteristics of boundedness.Laser scanning technology has the characteristics of high sampling rate,high resolution,high precision and panoramic scanning.However,the acquired indoor three-dimensional point clouds are usually unstructured and do not contain semantic information.Indoor mapping and three-dimensional modeling using laser measurement are faced with a series of challenges: due to the systematic errors of measuring equipment and the complexity of indoor environment,the collected three-dimensional point clouds usually contain varying degrees of noise.The indoor space contains a series of furniture and facilities,the clutter and occlusion in indoor environment results in data missing and holes in the acquired 3D point clouds.Therefore,the study of only extracting structural element(walls,ceilings and floors)from point clouds can not build a complete building information model.Some studies focus on creating visually realistic models,which do not include semantic information and can not meet the needs of urban management,architectural design and simulation.Considering the cellular nature(i.e.,rooms)of indoor spaces,recent studies have often reconstructed 3D indoor models from point clouds in heavily occluded environments by room space decomposition.However,recent space decomposition methods have excluded connection spaces(e.g.,doors)and neglected the fact that room spaces are connected by connection spaces.These methods have thus failed to determine the correct room spaces in the case of unclear wall constraints from point clouds.The modeling of staircase connected space becomes a difficult problem when these methods are extended to multi-storey space.In this study,we first analysis the mapping relationship from indoor point cloud to interior building model,then put forward the logical model,namely semantic decomposition and recognition of indoor spaces with structural constraints,for 3D building modeling from indoor 3D point clouds.The semantic classification system of indoor space and the types of structural constraints are propsed.The concept of composite space is put forward,and the process description system of dividing indoor space into celluar space is improved.In order to extract the planar structural elements of large-scale indoor point clouds,an improved RANSAC method based on Normal Distribution Transformation(NDT)cells,called NDT-RANSAC,is proposed for three-dimensional point cloud plane segmentation.First,the three-dimensional point clouds are represented by discrete voxels,and the point sets in each voxel are described by normal distribution.The NDT feaures of each voxel,i.e.the digital characteristics corresponding to the normal distribution,are calculated.Each NDT cell is classified as planar and non-planar cell base on the calculated NDT features.A planar NDT cell is selected as a minimal sample in each iteration to ensure the correctness of sampling on the same plane surface.In order to extract the line features of point clouds in indoor environment with curved walls,an improved Mean-shift based line feature extraction algorithm is proposed for slicing point cloud,which realizes the extraction of wall curve features.It contains a novel straight and curved line tracking method followed by a straight line test.Robust parameters are used,and a novel straight line regularization method is achieved using constrained least squares.Through the above methods,the extraction efficiency of point cloud geometric elements is effectively improved.On the basis of extracting structural geometric elements of three-dimensional point clouds,a new room space decomposition method is proposed for single-storey three-dimensional point clouds.This method uses wall plane or wall curve as constraints,and uses morphological method to obtain initial room segmentation on free-space raster.The method constructs a cell complex with both straight lines and curved lines,and the indoor reconstruction is transformed into a labeling problem that is solved based on a novel Markov Random Field formulation.The optimal labeling is found by minimizing an energy function by applying graph cut approach.In this way,the accuracy of room space descomposition is improved.Through room space descomposition and room information extraction,room reconstruction is realized.The point clouds on the wall are segmented to achieve the wall reconstruction.A method of identifying and modeling the wall connection space is proposed.That is to say,the U-V direction region growing algorithm is used to extract the structural elements of doors and windows.The wall solid model is constructed by classifying the inner and outer walls,merging the primary and secondary walls and extracting the central line of the wall.Further,the method is extended to multi-storey environment,and the methods of floor partition based on clustering method and stair reconstruction based on spatial connectivity(stair connection space)are proposed.Finally,the method of building surface model and BIM model is studied.The integration and output of different types of models are realized,and the object-oriented three-dimensional model of indoor building with rich semantic information is generated.By connecting the room space and connection space with the topological constraints,the topological consistency of the generated three-dimensional indoor model is ensured.The method proposed in this study realizes three levels of building interior three-dimensional modeling: room to single-floor to multi-floor,room to connection space and surface model to solid model.These methods are validated on several synthetic and real world datasets.The results show that the proposed method has the advantages of accuracy and reliability,and is suitable for multi-storey environment with curved walls.