Research On Mobile Mapping Object Classification And Reconstruction

Compared to traditional spatial information acquisition technology,mobile measurement technology can provide high precision and high density three-dimensional coordinate information for target model reconstruction with the characteristic of initiative,non-contact,all-weather and high efficiency.The data processing technology of mobile measurement is very necessary and urgent,which has become the biggest bottleneck restricting the further application of mobile measurement technology.In order to improve data processing efficiency of mobile measurement in urban survey,this paper studied urban objects classification and information extraction of vehicle point cloud and panoramic image,including the technology of efficient management and scheduling,hierarchical segmentation,and obj ect-oriented classification of huge point cloud data.Besides,the fast mapping and 3D reconstruction of city scene based on the classification of point cloud data are also studied in this paper,which provide a reference for automatic mapping and modelling of urban scene using mobile measurement technology.The main contents of this paper are as follows:This paper studied the efficient management and scheduling technology of mobile measurement data,designed index data structure for track,point cloud and panoramic image respectively,and based on which,realized the integration of multi-source data,it can quickly locating point cloud and panoramic data through track data,realized the high precision registration of point cloud and panoramic data,as well as point cloud coloring,which providing basic data structure for subsequent segmentation and classification.This paper studied the segmentation of point cloud.A hierarchical segmentation strategy which selects optimal segmentation method for each object from easy to hard is proposed based on the analysis of traditional segmentation method of point cloud.With this strategy,the ground,buildings,pole-like objects and so on can be segmented from urban scan scene efficiently,providing a foundation for further classification and extraction processing.Based on the point cloud segmentation,this paper studied object-oriented classifications of pole-like objects,including feature extraction and selection of pole-like objects,the SVM,random forest,and convolutional neural network classification methods.It proposed an effective point cloud classification method based on convolutional neural network which realized the feature self-learning of point cloud with deep learning.This paper studied the object extraction method combining image data and point cloud data,aiming at solve the problem that some special objects are hardly extracted from point cloud data merely.It studied the object extraction algorithm based on image,discussed two improved algorithms to detect objects,then used point cloud depth image to obtain their absolute coordinates,and finally realized the object information extraction.Based on the segmentation and classification of point cloud data,the automatic achievements transformation method for scanning point clouds of different urban targets is studied,and a multi-level intelligent object information extraction and reconstruction strategy is proposed.The processing methods of different objects in digital mapping and 3D modeling are discussed in detail,and the corresponding flow chart is given.On the basis of the above researches,this paper designed software of vehicle point cloud data processing for geomatics application,creating a system platform base for the promotion of the project application of mobile measurement technology.