Point Cloud Segmentation Based On Improved Regional Growth And Cost Optimization

With the continuous development of 3D scanning equipment,point cloud segmentation has become a research hotspot in the field of machine vision.As an important step of shape classification,object detection and tracking,point cloud segmentation has been the focus of many researchers in recent years.In the fields of urban governance and remote sensing mapping,point clouds are used as input data for 3D scene modeling,and the results of point cloud segmentation have a great influence on the subsequent building reconstruction process.At present,the point cloud segmentation method for the roof of a building still has insufficient accuracy,and the segmentation effect of the transition area of the roof boundary is poor.In response to the above problems,this paper proposes a method that combines improved area growth and global energy function optimization.This algorithm can achieve accurate segmentation of building roofs and effectively improve the segmentation results of the transition area of building roof boundaries.The main research work of the thesis is as follows:Region growth is a very efficient point cloud segmentation method,but it has problems such as poor segmentation of boundaries and transition regions.Based on this,this paper proposes a new boundary voting optimization algorithm based on the initial segmentation results of region growth.The algorithm can re-vote some unassigned isolated points and erroneous segments in the boundaries and transition regions,and can adaptively converge on the boundaries of different segmentation results.In addition,because region growth can cause problems such as under-segmentation and over-segmentation,this algorithm can effectively improve the number of over-segmentation.At the same time,the algorithm can also eliminate some erroneous facet segments in the region growth result,avoid the expansion of invalid labels in the subsequent energy function optimization process based on graph cuts,and improve the optimization efficiency of the energy function to a certain extent.The method based on global energy function optimization is used to further optimize the initial segmentation results,and the energy function is constructed for the point cloud segmentation problem of building roofs,which includes data cost,smoothing cost and label cost,and the graph cut method is used for optimization.By establishing a corresponding graph model for the energy function,and assigning certain weights to each edge in the graph,the maximum flow/minimum cut algorithm is used to solve it.Through continuous iterative optimization,the local minimum of the energy function is finally obtained.Finally,this paper conducts experiments on the standard data set collected by lidar scanning equipment,and compares several building roof point cloud segmentation methods.The experiment shows that the method proposed in this paper can be used for the transition area of multiple roof planes.And the boundary achieves a good segmentation effect,and can significantly improve the accuracy of the point cloud segmentation of the building roof.