Research On 3D Point Cloud Denoising Alorithm Based On Bilateral Filter

With the rapid development of computer science and 3D scanning technology,the application of point cloud data is more and more extensive.However,in the process of obtaining point cloud data,due to the impact of external environment,human operation,object material and equipment accuracy,the acquired point cloud model often mixes with noise,which seriously damages the geometric characteristics of the model,and brings adverse effects to the model packaging and detection.Therefore,in order to obtain a better3 D model,it is necessary to denoise the point cloud data.In view of the problem that it is difficult to remove the noise of different scales at the same time in the process of point cloud denoising,combined with the existing denoising algorithm,this article divides the point cloud noise into large-scale noise and small-scale noise,and processes the noise of different scales step by step.Large scale noise refers to outlier noise,including outliers far away from the main body of point cloud and small and dense noise points;small scale noise refers to the noise points mixed with the main body of point cloud.First of all,for the problem that two kinds of outliers cannot be removed at the same time in large-scale noise,this article adopts the method of combining radius filtering and dbsan clustering to remove large-scale noise.Radius filtering is used to remove outliers far from the main body of point cloud,DBSCAN clustering algorithm is used to remove small and dense noise points far from the main body of point cloud.The parameter selection methods of radius filtering and DBSCAN clustering algorithm are proposed respectively,which can reduce the manual parameter adjustment and realize the adaptive parameter selection.Secondly,after removing large-scale noise,the main body of point cloud data is confused with non outlier noise,namely small-scale noise.For this kind of noise,this article proposes an adaptive bilateral filtering algorithm based on point cloud feature classification.In the aspect of point cloud feature classification,the point cloud is divided into flat region and sharp region by using normal distance and average curvature double thresholds,which improves the robustness of feature classification.In the estimation of normal vector,the adaptive normal estimation method is used for sharp region,and the principal component analysis method is still used for flat region,which can not only keep sharp characteristics,but also avoid the problem of the decrease of calculation efficiency caused by the adaptive.Finally,in the aspect of bilateral filtering,different spatial weights are used for sharp region and flat region,and adjustment function is introduced for feature factor to realize adaptive bilateral filtering for point cloud data.Finally,the algorithm in this article is analyzed.Experimental results show that the algorithm can effectively remove the noise of different scales.The outlier denoising algorithm proposed in this article can simultaneously remove outliers and small and dense noise points.Compared with other algorithms,the point cloud data after denoising is the closest to the original model.The adaptive bilateral filtering algorithm keeps the sharp features of the point cloud while denoising,avoids excessive fairing,and reduces the standard deviation of the model after noise removal,which effectively improves the denoising accuracy of the point cloud.