Research On Multi View 3D Reconstruction Method Based On Constrained Delaunay Triangle

In recent years,the demand for urban-level real-time 3D model has become more and more urgent,and the rapid acquisition of high-quality building model has become a hotspot in the field of computer vision and graphics.Based on the three-dimensional reconstruction method of Multi-view Stereo(MVS),it is the first choice for 3D reconstruction of large-scale scene because of its low cost of data acquisition and real texture and process automation.Program.However,with the application of various types of three-dimensional model to improve the quality requirements,the current three-dimensional reconstruction of MVS there are some shortcomings need to be improved.On the one hand,the MVS method needs to match the pixel-to-pixel similarity,and the surface of the urban building is usually lack of texture,resulting in a matching failure or error,which causes the quality of the reconstructed model to degrade,distort the distortion,and the unevenness of the texture On the other hand,for high-resolution aerial images,pixel-by-pixel dense matching process,directly lead to the generation of cloud inefficient,difficult to apply to a wide range of urban scene three-dimensional reconstruction.In order to solve the above problems,this paper proposes a 3D reconstruction method based on constrained Delaunay triangles and a fast method based on PactchMatch to generate point cloud and improve image-based MVS.It aims to improve the modeling accuracy and speed up the modeling efficiency.The main work of this paper is summarized as follows:1)proposed a three-dimensional reconstruction method of urban architecture with linear constraints.The basic flow of the method is as follows:Using a series of pictures obtained by the UAV as input,the straight line and edge of the image are first extracted by the Line Segments Detector(LSD)and the super pixel,and the edge polygon is simplified,Constrainted Delaunay Triangle(CDT)to obtain a single view 2D triangular grid,and then according to the dense point cloud back to the 2D triangular mesh vertex and three-dimensional position get a single view of the 3D grid model,and finally through the multi-view merger to get the complete scene 3D model.2)This paper proposes an improved method of generating point cloud,which mainly improves the speed of cloud generation to speed up the modeling process.The basic flow of the method is as follows:Obtain a set of pictures using the UAV,through the sparse Census match to obtain the depth map,the depth map is estimated,filtered,fused and backes to get 3D point cloud data.Finally,the scene of multi-angle aerial images of real-time scene acquired by UAVs is used to model the urban scene.Experiments show that the method has good robustness and stability,and the final three-dimensional model is of high quality and fast generation.