Research On De-noising Algorithm Of Point Cloud Model That Preserves Features

As real as possible virtual scene will increase the students' interest of study and strengthen the effect of education in educational games and virtual teaching system.So,there is a huge demand for realistic digital three-dimentional(3D)models.Simply constructing these models manually in 3D software is time-consuming and energy-consuming.However,with the fast development of 3D scanning device and technology,it is possible for us to rapidly digitalize real objects in the world.Through 3D scanning device,the surface information of real objects can be transformed and stored in point cloud,which is a kind of data struct storing point information.We can reconstruct the point cloud and obtain the mesh model that can be easily stored?rendered and controlled.However,the technology and method about acquiring and handling point data still have a lot of limitations.Especially,because of the bad effects of artificial disturbance?scanning device limit?scanned model defect and environmental impact,the point clouds acquired by 3D scanning device will suffer several problems such as non-uniform?noisy?having outlier and so on.But the point cloud model we need is high-quality when we execute some further operation like reconstructing.Recently,there are many researchers focus on point cloud denoising.Due to the diversity of 3D geometry model and complexity of noises,it is still a challenge to robustly removing noise while maximally preserving sharp geometric features.The main work of this paper include the following aspects:1.We do some researches on the basic point cloud denoising.We first introduce some geometry operations for point cloud processing,for example,computing point neighbor,calculating point cloud normal and point cloud normal filtering.We also illustrate the details of locally optimal projection,which is a classic point cloud denoising method.The main idea of this method is also explained in detail.2.Via analyzing the locally optimal projection method,we present a feature-preserving anisotropic weighed locally optimal projection denoising method for point cloud.Setting a noisy model and its filtered normal as input,this method we designed can output a clean?uniform-distributed?feature-preserving point cloud.3.We introduce a kind of probabilistic model——Gaussian mixture model.Based on this model,we present a new anisotropic point cloud denoising method.This method also need the filtered point cloud normal as input and it is feature-preserving.4.We compare the two denosing method presented in this paper with four state-of-the-art method,including visual comparison?quantitative comparison and the running time.The further research about the theory differences between the six methods is also explored in our paper.