Denoising Algorithms In 3D Point Cloud Reconstruction

In recent years,with the development of reverse engineering and 3D scanning technology,3D point cloud processing technology has received extensive attention and application.In the actual application process,3D point cloud reconstruction is generally to reconstruct the model by acquiring 3D point cloud data from physical objects.In the process of acquiring point cloud data through 3D scanning instrument scanning or image matching reconstruction,the target point cloud is acquired due to factors such as ambient light,measurement equipment accuracy,surface material,human error,stereo image calibration error and matching reconstruction error,etc.The data always contain deviation points and error points,the deviation points and error points are called noise points.These noise point cloud data bring a lot of trouble to subsequent point cloud registration,segmentation,reconstruction,high-precision measurement and other processing.Therefore,point cloud denoising is crucial in the reconstruction of the entire point cloud model.For these noise point cloud data,combined with the existing denoising algorithm,the large-scale outlier noise cannot simultaneously remove the sparse noise and isolated noise caused by sampling unevenness,and the point cloud model with complex geometric features cannot maintain point cloud detail and edge features when denoising.In this paper,the noise point cloud is divided into outlier noise points that are not mixed with the point cloud body and non-outlier noise points related to the point cloud body for denoising.Firstly,according to the distribution characteristics of outlier noise points,this paper proposes a statistical feature analysis method based on the local neighborhood distance distribution of noise point clouds to remove the sparse outlier noise,then remove the isolated outliers according to the density within the optimal neighborhood radius.The method solves the problem that sparse noise caused by sampling unevenness and some small and dense isolated noise cannot be removed at the same time.Secondly,for the point cloud after removing the outlier noise points,there are still some edge burrs and noise points floating on the surface of the point cloud body,which are called non-outlier noise points.for these noise points,this paper proposes an anti-noise robust normal estimation algorithm,which is used to estimate the normal and curvature of the noise point cloud to detect the sharp features of the noise point and the geometrically complex feature points.Finally,the noise point cloud is divided into different regions according to different geometric complexity.The adaptive bilateral filtering algorithm based on the improved point cloud curvature is used to remove the non-outlier noise points to maintain the edge and detail features in the noise point cloud denoising process,and to prevent edge features from smoothing and volume deformation of the point cloud model.In order to verify the effectiveness of the proposed method,this thesis conducts experiments on several different types of point cloud data,and compares them with the existing denoising algorithms.Through the analysis of experimental results,compared with other algorithms,the proposed algorithm can remove different types of noise from outliers and non-outliers,and also maintain sharp features better such as point cloud edges and details,and improve the denoising accuracy of point cloud.