The Research And Realization Of The Integrated Display Technology Of 3D Terrain Scene

In recent years,information data extraction of remote sensing images and reconstruction and rendering of three-dimensional scenes have received extensive attention from researchers.Since the color difference between the ground object and the surrounding background is different,the use of a single method to extract the information data will result in low accuracy or low inefficiency,which cannot meet the demand.Therefore,this thesis proposes two methods to extract the remote sensing image feature information data for the two cases of color difference between the ground object and the background.When there are few remote sensing images and there is a large difference between the color of the extracted feature and the background color,the information extraction methods requiring large samples cannot accurately extract the information.Therefore,this thesis uses fuzzy c-means and support vector machine algorithms to get classify pixels and feature pixels.Then we use Bwareopen and mathematical morphology methods to denoise the image and then get the collection of feature pixels.When there is little difference between the ground object color and the background color and there is a large amount of remote sensing image data to be trained,the image data extracted by some machine learning method has more noise.In view of this situation,this thesis proposes a pixel segmentation method based on the full convolution residual network model.This method first uses the full convolution residual network to learn the characteristics of the dataset from a large number of remote sensing image samples,and then we obtains a network model that can detect remote sensing image features by training.Finally,the model is used to divide the image into ground objects and non-ground objects.In addition,for reconstruction and rendering problems,the reconstruction and rendering in traditional three-dimensional scenes use delaunay triangulation(the segmentation-merge)algorithm.This algorithm has a high time complexity and cannot meet the real-time requirements when the terrain data volume is large.Therefore,based on the control point data of feature information extracted by full convolution residual network algorithm,this thesis proposes a fast method for three-dimensional scene construction and multi-level rendering.Firstly,the method uses the cubic B-spline curve function to fit the center points to the centerline of the ground object.Then the terrain data is modified according to the multi-resolution level of the terrain,and the ground object model is reconstructed.Finally,the terrain resolution hierarchy is used to perform multi-level rendering and we achieve long-distance blurring and close-distance visibility.