Research On Semantic Segmentation Method Of Building Facade Point Cloud Based On IMMS

In recent years,the application of indoor three-dimensional architectural models has become more and more extensive,such as digital city,BIM,indoor navigation and so on.The use of various instrument platforms to obtain laser radar point clouds data to construct real three-dimensional models has become an important means of obtaining indoor architectural models.At the same time,many scholars have researched how to extract and separate different architectural elements from point clouds data.However,for different instrument platforms,different scenarios,and different point clouds data sizes,it is still very difficult to automatically segment architectural elements(such as the ground,windows,surrounding walls,doors,etc.).In this paper,the indoor mobile measurement system(Indoor Mobile Measurement System-IMMS 3D)is used as the data source.On the basis of collecting the indoor data of the building,the emphasis of this paper is on the semantic segmentation of the interior Point cloud of the building.The main work and achievements of this article are as follows:(1)Ground-based laser scanners are often used in the acquisition phase of buildings,and this article uses the indoor mobile measurement system(IMMS 3D)that appears on the market as the experimental data source.The data types,characteristics,and composition of the point cloud data collected by the system are analyzed,and the advantages and disadvantages of the system are analyzed.The point cloud data collected by the system and the points collected by a traditional ground laser scanner are compared.The difference in cloud data and the reasons for using the system as a source of data.(2)According to the diversified features of building interior architecture elements,the idea of“first segmentation and then classification" is adopted.First,an improved RGB_I region growth algorithm is proposed for the architectural elements with conspicuous convexity and the segmentation is achieved.The over-segmentation of clouds,for over-segmented point cloud data,proposes an improved multi-conditionals European clustering method to perform secondary segmentation and gradually segment the point cloud data of different architectural elements.(3)For the case where the semantic information of point cloud data is difficult to judge,the idea of“reducing complexity and simplifying" is applied.The outward bounding box is added to the segmented point cloud block twice and the geometric information of the external envelope is calculated.And with the a priori geometric information of architectural elements want to be compared,to give point cloud block semantic information.The judgment of the semantic information of the point cloud is converted into the judgment of the geometric information,and the semantic segmentation is finally completed.