Research On Indoor Semantic Map Construction And Navigation Method Based On 3D Laser Point Cloud

As urbanization progresses,the interior scale of buildings becomes larger and larger,and the indoor environment becomes more and more complex,which makes indoor route finding and route planning more difficult,thus giving rise to the urgent need for indoor navigation.As the basis of indoor navigation,indoor semantic map construction and navigation methods are important tools for indoor spatial cognition,and gradually become a hot research topic in the field of indoor navigation.The traditional indoor map construction and navigation method research takes static scenes as the design premise and builds them with data such as 2D floor plans of buildings,BIM and City GML,but the indoor maps generated with traditional data will deviate from the real environment,and are difficult to obtain and not easy to update.Compared with traditional data,point cloud data can better reflect the geometric information and spatial relationship of objects,and gradually becomes an important data source for indoor semantic map and navigation network construction.In this thesis,we select point cloud as the data source and propose a point cloud-based indoor semantic map and navigation network construction method.Semantic segmentation of interior element point clouds is achieved by improving Point Net++ network.Design a point cloud-based indoor semantic map construction method to divide indoor elements into doors,walls,cartographic elements and non-cartographic elements,and project them onto a two-dimensional plane to build an indoor semantic map;construct a topological navigation network for indoor scenes based on the topological relationship between indoor elements and Indoor GML standard,and optimize the navigation paths by combining the indoor semantic map.The main research of this thesis is as follows:(1)A semantic segmentation method for indoor field attraction clouds is studied.To obtain the semantic information of indoor elements,a Point Net++-KPConv point cloud semantic segmentation model is designed with Point Net++ network as the base network.The original loss Loss1 is obtained by Point Net++,and the output of Point Net++ is processed using the KPConv module to obtain Loss2,and Loss1 and Loss2 are grouped and summed to form the loss of the Point Net++-KPConv network.the introduction of the KPConv module can make the model pay more attention to the local features of the point cloud and thus improve the accuracy of point cloud semantic segmentation.(2)An indoor semantic map construction method was studied.A point cloud-based indoor semantic map construction method is proposed for the problem that movable obstacles in the scene are not considered in the construction process of existing indoor semantic maps.By identifying the semantic information of the point cloud of indoor elements with convolutional neural network,and discriminating the obstacle types based on the semantic information and point cloud features,the indoor elements are divided into doors,walls,cartographic elements and non-cartographic elements,and the results are projected onto a two-dimensional plane to build the indoor semantic map.(3)The indoor navigation network construction method was studied.To address the problem that the current path planning only considers indoor spatial layout and ignores movable obstacles,the indoor navigation network is constructed according to Indoor GML standard by defining the topological relationship between indoor elements;it is optimized for the limitations of the navigation network combined with indoor semantic maps;and the Dijkstra algorithm is used to realize the path planning based on indoor navigation network for the needs of indoor navigation.In this thesis,the proposed method is experimentally validated and analyzed in S3 DIS and Matterport3 D datasets.The experimental results show that o Acc,Acc and m Io U of Point Net++KPconv model are better than Point Net,Point Net++ and KPConv-FCNN models;the indoor semantic map and navigation network constructed by the indoor semantic map construction and navigation method can accurately represent the indoor environment,which provides an important reference for the research and application in the field of indoor navigation The indoor semantic map and navigation network constructed by the indoor semantic map construction and navigation method can accurately represent the indoor environment and provide important reference and support for research and application in the field of indoor navigation.