Research On Three-dimensional Point Cloud Reconstruction With Adaptive Segmentation And Curve-driven Topology

The three-dimensional reconstruction can reconstruct three-dimensional objects based on the scanned data and provide a new way for people to observe the objects from different perspectives.With the development of virtual technology,people are pursuing higher precision in digitization and visualization.In terms of 3D point cloud processing and reconstruction,there are still some problems,such as incomplete point cloud data,large data,and low reconstruction accuracy.In order to address these problems,this paper proposes a preprocessing and segmentation algorithm for 3D point cloud,and an interactive surface reconstruction algorithm based on segmentation and curve-driven topology to implement 3D reconstruction and improve the precision.The main research content and conclusion of the paper are as follows:(1)Point cloud preprocessing method based on feature-enhancement.Aiming at the problem of the original scanned point cloud missing a large amount of detailed information,large data,and low efficiency,a point cloud preprocessing method based on feature-enhancement was proposed to address these problems.The method includes three steps: First,a point cloud simplification algorithm based on adaptive local projection operator is used to simplify large-scale point cloud;then aiming at the problem of normal vector is sensitive to noise,the normal vector based on anisotropic neighbor search is used to optimize the normal vector and reduce the calculation error of the normal vector;finally,aiming at the problem of some details are lost after simplification,an edge-aware feature enhancement algorithm is proposed to add points along the sharp parts and edges to increase detail information.The experimental results show that the model can preserve the shape features of the original data when the reduction rate reaches above 95%.(2)A self-adaptive segmentation method for a point cloud.In order to solve the problem of different reconstruction accuracy in different parts of an object,this paper proposes an adaptive point cloud segmentation algorithm.This algorithm automatically selects cluster centers according to the extracted features and segments point cloud according to the cluster centers.It divides the point cloud into different areas with similar structures,which can realize effective segmentation of the point cloud.Experimental results show that the accuracy of point cloud segmentation is increased by 10.2% when compared with other segmentation algorithms such as Rand Cuts and Shape Diam et al.In order to reduce the time of extraction feature,the method of point cloud segmentation based on OpenCL is proposed,which optimizes the data storage structure and improves data access efficiency according to the parallel architecture and hardware characteristics of the GPU.Experimental results show that the efficiency of point cloud segmentation based on OpenCL is 22 times faster than that of the CPU.(3)Interactive surface reconstruction algorithm with segmentation and curve-driven.Based on the results of preprocessing and segmentation,an interactive surface reconstruction algorithm with segmentation and curve-driven is proposed.First,the point cloud skeleton curves are extracted according to the segmentation results;and then the profile curves of the surface are formed according to the extracted skeleton curve;finally,all the generated profile curves are optimized to fit the point cloud to complete the three-dimensional model reconstruction.The experimental results show that the reconstructed results of this algorithm can maintain the details and shape features of the point cloud,the error is lower than 0.1 and the precision is high,in addition,the reconstruction results can be faithful to the morphological characteristics of the model.