Research On Feature Point Extraction Method Of 3D Discrete Point Cloud

With the popularity of computing resources and 3D scanning hardware,the processing speed of point cloud data is faster and faster,and the acquisition of point cloud data is easier and easier,which makes the application of point cloud more and more widely.Point cloud has been widely used in many fields,such as reverse engineering,cultural relics restoration,architecture and so on,and point cloud feature extraction is an important step in the use of point cloud data.With the improvement of the quality of life and the progress of technology,various fields have higher and higher requirements for the accuracy and speed of point cloud feature extraction.Because of the breadth of application sectors,point cloud data becomes quite complex.The obtained point cloud data frequently has noise and non-uniformity due to the surroundings and equipment.In addition to external considerations,the form of point cloud data differs substantially among sectors.Some items in some sectors,for example,are dominated by sharp characteristics,whilst others are dominated by smooth features.Point cloud feature extraction is a critical stage in the process of denoising,matching,segmentation,and identification of point cloud data.The quality of point cloud feature extraction has a significant impact on the outcome of later point cloud processing.As a result,the focus and challenge of point cloud feature extraction is rapid and accurate feature extraction in a noisy environment.Many academics have conducted extensive study on point cloud feature extraction in the past,however there is no universal approach that can be used to point cloud feature extraction in all cases.This thesis studies the feature point extraction and extraction efficiency of point cloud.This document mostly contains the following work contents:(1)In order to solve the problem of incomplete feature extraction in the part where the surface change is not obvious,a point cloud feature point extraction method based on the combination of surface change and density is proposed.The surface variation is measured using Laplace shrinkage,which has a high noise resilience.However,while this approach performs well for sharp feature extraction,it performs poorly for smooth feature extraction,resulting in inadequate feature extraction.The feature component of some point cloud models has more dense points.The impact of density is taken into account in the surface change approach,and the contribution values of two variables are established by assigning distinct weights.Observe the effect of shrinkage or density on feature extraction using several models to change the weight,and eventually develop a general assessment standard.The experimental results show that the model with dense points in the feature region can achieve more complete feature extraction results through the combination of the two methods.(2)An octree-based partition extraction approach is proposed.Because general point cloud model data is big,feature extraction takes time.The computing efficiency is much worse when calculating numerous indicators.To boost efficiency,the octree approach is used to minimize the calculation scale.Specifically,the first step is to carry out rough calculation,exclude points that are obviously not characteristic areas,and obtain a relatively small number of points for fine calculation.Firstly,the octree idea is adopted to subdivide the model layer by layer until the lattice side length reaches the set value.Then the characteristic index of each nearest point to the lattice center is calculated.If the index is greater than the threshold value,the nearest point to the lattice center is taken as the center,and the search radius is 2.5 times of the length of the crystal lattice side.Then all the points in the radius neighborhood are taken as candidate points for the second judgment.The purpose is to add the points that may be feature points to the coarse extraction point set,to ensure that the result of feature extraction is connected to avoid missing judgment,and to ensure that there are enough neighborhood points in the fine extraction stage.The second step is the fine extraction stage.All the points extracted in the first step are taken out to calculate the feature extraction index,so as to realize the extraction of the final feature points.Experimental results show that this method can significantly improve the efficiency for simple models.