Applied Research On 3D Reconstruction Of Urban Building Based On MVS Point Cloud

In recent ten years,three-dimensional(3D)reconstruction technology based on image has developed rapidly,it has obtained the good application effect in some fields.As its basic technology,the Multi-view Stereo(MVS)mainly have three steps:To calculate the sparse 3D information of camera and scene by SFM(Structure from Motion),get 3D point cloud model by MVS dense matching,and then generate surface model from point cloud.This paper focuses on the last step,taking the 3D reconstruction of urban building as the goal,in order to study the surface reconstruction based on MVS point cloud.Previously,3D reconstruction from large-scale urban scenes mostly based on laser radar(LiDAR)point cloud,which has a good accuracy,but the collection of LiDAR data is difficult and costly.The MVS point cloud is rich in image information,low-cost and relatively easy to obtain.However,because the image matching is not accurate enough and other reasons,the MVS point cloud is noisy and miscellaneous.The methods of surface reconstruction based on LiDAR point cloud are not applicable to the MVS point cloud.Therefore,the study on the approach of 3D reconstruction based on MVS point cloud has become urgent.Meanwhile,in terms of the 3D modeling of urban building,due to the diversity and complexity of building,how to generate a true reflection of building geometry information during the surface reconstruction stage,not only to overcome the impact of the noise existing in MYS point cloud,but also avoid the excessive smoothing in the outline of building,is a core task and difficulty.Aiming above problems,in order to improve the quality of urban 3D model,this paper proposes a optimized method of urban building surface model based on MVS point cloud,The main work is summarized as follows:1)Research on the automatic building extraction from point cloud covering complex urban scenes,and a process for automatic extraction of buildings based on MVS point cloud was put forward.The process is simple and efficient,and then,through the use of projection,morphological dilation,contour extraction technology,finishing the automatic building extraction of MVS dense point cloud covering complex urban scenes which includes roads,vegetation and so on was implemented..2)In terms of the inadequate of Poisson surface reconstruction in urban 3D modeling,a optimization method of surface reconstruction based on RANSAC fast fitting is proposed.This method firstly makes a denoising on the building MVS point cloud,then applies the RANSAC fast fitting method based on block to the filtered point cloud.Finally,the optimized surface model generated by the Poisson surface reconstruction.Through experimental analysis,the plane and sharp features of optimized surface model can be effectively restored,and the geometric information of optimized surface model is more close to the real model.The article finally concludes a work process of urban building 3D reconstruction based on MVS point cloud.By applying the method to the large urban MVS point cloud,we finish the reconstruction of urban building covering large scenes,and prove its good applicability and ability of reconstruction.