Research On 3D Mesh Model Reconstruction And Refinement Driven By The UAV Image Information

3D(Three-Dimensional)model is a visualuation for 3D spatial information.It can reproduce the 3D real world scene more realistically and accurately in computer environment.The components of the 3D model reconstructed by the multi-view images are grid and texture.The 3D Mesh model composed of grid point cloud generated by the object,and texture is the grooves of objects' surface.A 3D model is formed by mapping the texture to the surface of an object in a specific way.The existing 3D reconstruction method based on the multi-view images cannot have good visual effects,it restricts the development of 3D model in a certain extent.The existing reconstruction problems has a bad influence on 3D model result.In image data acquisition,it adopts UAV photogrammetry technology,which caused delay between the common UAV shooting recording camera exposure time and the real camera exposure time;And in dense matching,the existing dense matching method has a worse result and slower speed;Morever,the 3D Mesh model constructed by the 3D point cloud using existing methods is an approximate surface,which exists some problems of too dense 3D Meshnon-manifold vertices and edges,and so on.This paper is mainly research on 3D Mesh model reconstruction and refinement driven by the UAV image information,to solve the existing problems of 3D model reconstruction based on multi-view images.Aiming at the low accuracy problem of aerial space triangulation,we adopts the GPS-supported bundle adjustment to improve the accuracy of aerial space triangulation,and obtain the higher accuracy of interior and exterior orientation elements;With slower speed and worse result of dense matching,we adopts SfM technology to obtain the 3D sparse point cloud,combines PatchMatch and NCC,to densify the 3D sparse point cloud,improve dense matching effect and dense matching speed to obtain the high accuracy of 3D dense point cloud;With the existing problems of constructing 3D Mesh model based on the point cloud,we modify the Graph-Cut to construct the manifold 3D Mesh model;Then processing the Mesh model to obtain the simplified and subdivided 3D Mesh model;Finally,we take the images data as the drive information,adopts image maching method between images,to refine the Mesh model to obtain the higher accuracy of 3D Mesh model.The 3D Mesh model has a a smallest error with the real surface model,so as to to improve the accuracy of the final reconstructed 3D model.The main contents and innovations of this paper are as follows:1)In the section of three arieal triangulation: When UAV obtained images,it has caused delay between the common UAV shooting recording camera exposure time and the real camera exposure time.Therefore we adopts GPS-supported bundle adjustment to solve the aerial space triangulation,which improved the final adjustment results and obtain the higher accuracy of image interior and exterior orienataion elements.2)In the secion of dense matching: In the process of finding the same point in stereo images,because the image information has large amount of calculation,memory consumption and so on.Therefore,we adopts the SfM to reconstruct the 3D sparse point cloud,combines the PatchMatch and NCC to densify the sparse point cloud,which can improve dense matching speed and dense matching result,to obtain the higher accuracy of 3D dense point cloud.3)In the section of Mesh model construction: We modify the Graph-Cut,constructing 3D Mesh model based on the point cloud,to construct the manifold 3D Mesh model.Moreover,we process the 3D Mesh model which exists some problems of too dense 3D Mesh,local deformation,not prominent edges and corners,non-manifold vertices and edges,isolated components,isolate islands,small areas,holes,spurious components,and so on,to obtain the simplified and subdivided 3D Mesh model.4)In the section of 3D Mesh model refinement: With the driven information of images data and image matching method between images,we re-project an image to other visible images induced by the object triangular facet,adopt gradient descent method,take the derivative of object vertex coordinates with the images' correlation coefficient to obtain the gradient change value.With intersection angle between the image to take weighted average to gradient chage value to obtain a new gradient change value and adjust the vertices coordinates of object 3D Mesh in order to minimize regional matching cost,to obtain the refined 3D Mesh model.