Research On Multi-resolution 3D Terrain Modeling Technology Based On Terrain Semantic Segmentation

Terrain model is the integration and reduction of terrain features.It is a carrier and tool for people to represent terrain in three-dimensional space.The emergence of computers has enabled the digitization of terrain models.The Concept of "Digital Earth" has greatly promoted the development of 3D terrain modeling technology.However,the large increase in data volume has also brought new challenges to the current terrain modeling technology.How to quickly and accurately implement 3D terrain modeling has become one of the hot topics in current research.According to the actual application requirements in the field of 3D simulation,this paper studies the effective ways to improve the data redundancy and local distortion of the 3D terrain model,proposes a multi-resolution 3D terrain modeling method based on terrain semantic segmentation and completes modeling process to get terrain entity model.The test results of multiple sets of data samples show that the content studied in this paper can effectively improve the model data redundancy and local distortion problems.The main research work of this thesis includes the following aspects:(1)A terrain semantic extraction method based on extracting the extreme point of Gaussian curvature of hash point cloud is proposed.This method is based on the feature curve extraction method based on extracting the extreme point of Gaussian curvature,by setting the feature point extraction threshold of the algorithm,and combining with the hash point cloud curvature estimation algorithm to improve the algorithm feature area extraction effect.In order to achieve the purpose of algorithm improvement,this paper proposes a concept of"characteristic contour area",and uses this concept as a guide to complete the definition of the terrain semantics of mountainous terrain,and obtains "ridges","valleys","mountains","hills" "and" flat land "five kinds of semantic regions.And according to this definition,the extraction and segmentation of terrain semantic regions are realized.This semantic extraction method combines the discreteness of terrain data with the continuity of terrain semantics,which can achieve accurate segmentation of terrain semantic regions,which is beneficial to improve the accuracy of model construction.(2)This paper proposes a method for irregular segmentation of point cloud data based on terrain semantic segmentation.Based on the results of terrain semantic extraction,this method uses the semantic area as the basis for the point cloud data block and logically blocks the point cloud data.Different from the traditional rectangular boundary regular block method,this block method has the characteristics of irregular shape of the data block boundary,and the data amount of different data areas is not completely equal,which can well protect the terrain detail information and will not cause the occurrence of loss or distortion of details due to blocking.This method is helpful to improve the local distortion of the model in the current 3D modeling.(3)The multi-resolution distribution of terrain surface is realized by using the simplified technology of point cloud data classification based on semantic area.In this paper,based on the results of data segmentation,different data blocks correspond to different semantic characteristics,and different levels of data reduction rates are used to resample each data block to achieve simplified data classification.This work makes the difference in simplification rate between different semantic regions of the surface,thereby forming a multi-resolution distribution of the surface.And the hierarchical simplification can effectively reduce the number of point clouds in some semantic regions,filter out useless data,obviously improving the data redundancy existing in the current 3D modeling.