Research And Application On Point Cloud Preprocessing Based On Optimization Algorithm

In the process of establishing the grey model digital information platform,it's necessary to take advantage of reverse engineering technology to get 3D digital information from entity model.Reverse engineering is one of the most important technologies in the manufacturing field.Data collection,point cloud data preprocessing,3D model reconstruction is the main steps of reverse engineering.Point cloud data preprocessing is consisted of point cloud denoising,simplify,registration,feature recognition,region segmentation,geometric estimation.The denoising and simplification of point cloud data have a great influence on the quality of the subsequent 3D modeling,which is a very important step.Therefore,it has a certain scientific significance that research on point cloud data denoising and simplification optimization algorithm.The algorithm in this thesis is mainly applied to the Grey model's 3D reconstruction.Grey model is a kind of unique traditional architecture in LingNan area.It has a high cultural value and aesthetic value.In view of the near extinction of the grey plastic works,the establishment of the digital information platform to realize the inheritance and development of the gray plastic culture is of great significance.It's found that the point cloud data is scattered after analyzing the grey models.Scattered point cloud don't have any topological relation among points and points,and there is no general algorthim for denoising.The number of point cloud data acquired by 3D scanner is more than one million.There are noises due to machine and surrounding,which will seriously affect the quality of 3D reconstruction.What's more,it will consume more computer resources to store in and upload to the platform.In view of these difficulties,this thesis has made the following concrete research content on the basis of reading a certain number of home and abroad literatures:1.Based on the k-means clustering algorithm and the princple of cosine similarity,this thesis proposed a denoising algorithm based on k-means clustering.Determining the best clustering via the effect of k value on the denoise result.In each cluster,weigh the Euclidean distance by cosine similarity.It can strengthen the relationship between the boundary points and clustering center,avoiding the boundary points are mistaken for noise points.It can keep the boundary feature as well as identifying outliers.2.Based on the surface curvature and uniform grid,this thesis proposed an combined simplification algorithm based on curvature and uniform grid simplify.It use bounding box method to sort the scattered points,build KNN of every point,and calculate the curvature of the point.Simplify the point cloud according to 'the principle of curvature,and then resampling the removed points,using the method of the uniform grid to store the closest point far away from the centroid.The algorithm can be used to simplify the point cloud data and the details of the model are very good.3.In order to verify the rationality and validity of these algorithms,in the process of building information platform.The algorithms are applied to 3D reconstruction of grey model works by using the grey model works as the experimental object.It verified the feasibility and effectiveness of these algorithms,and get better results.The major innovation works are listed as follows:1.The establishment of point cloud denoising threshold,In order to avoid the boundary points are mistaken for noise points,joined the cosine similarity of Euclidean distance weighted,making more than the threshold further away from the cluster center,less than the threshold point is closer to the cluster center,thereby reducing the misjudgment of boundary points,to strengthen the efficiency of recognition of noise points.2.Aimed at the feature of grey model that more irregular surface,if only using the curvature to simplify will lead to producing blank in smoothing area.The algorithm in this thesis based on the curvature and the uniform grid,both to keep the feature points,and interpolation in a blank area.