Indoor Navigation Modeling Using RGB-D Sensor Data

With the increasing development of economy and technology,indoor navigation has become increasingly popular.Like outdoor navigation,indoor navigation also needs the support of navigation data.Due to the complexity of indoor structures,manual production of indoor navigation data is time-consuming and laboring,with relatively low efficiency.To address the problem raised from the state-of-the-art about indoor navigation,we in this paper develop a method of automatically constructing data files,which can be used for indoor navigation and location-based service based on indoor 3D point cloud data.The data model is constructed based on the IndoorGML data standard.The method proposed in this paper mainly includes the following three contents: the classification of indoor point clouds,the topology establishment among each indoor element and indoor navigation file construction.In this study we analyze the characteristics of the indoor point cloud acquired by Kinect sensor and present two classification methods.Based on whether there is motion trajectory data for the acquired data,this study proposes two methods of constructing the topology relationship.a)Point cloud classification is performed using Random forest classifier and then navigable elements information is extracted from the classification results.Firstly,color features and three-dimensional shape features are extracted,which will serve as the input of a random forest classifier.Finally,the classification results are optimized based on prior indoor knowledge.Experiments are conducted on the point cloud of teaching building and the position information and shape information are extracted from these classified elements.b)The spatial topology relationships among these extracted elements are established.In this section,we employed two methods to construct spatial topologies based on two types of data with different features.Because of the large degree of data overlap acquired by the Kinect sensor and the large registration error,we use sensor motion trajectory data to construct the topology of each space element.For the characteristics of the 2D-3D-S dataset and its lack of motion trajectory data,this paper adopts a spatial structure spilt based on the simple shape grammar to construct the spatial topologies.The experimental results show that the topological relationships among space elements can be effectively established.c)Navigation data files are generated for extracted spatial information.To organize the extracted spatial information and apply it to indoor navigation and location services,we use part of the contents of the Indoor GML data standard to build indoor spatial data files.The xml format data file can be quickly derived by the program language at the end.This study constructs the indoor topological data files which can be used for indoor navigation and location based services using indoor 3D point clouds.It concludes the information of states' position,topology relationship(such as: adjacency and connectivity)between spaces and each elements,mentioned in the standard of IndoorGML.The IndoorGML standard describes rich indoor information,however,the data file we construct only involved part of them.Therefore,we will study further to enrich the indoor navigation data information in the future.