Research On Mesh Reconstruction Based Feature Extraction Method On 3D Point Clouds

With the great progress of Computer Technology during recent years,the difference of machine perception and human perception gets smaller and smaller.Under this situation,the traditional 2D data is unable to meet the demand of representing 3D world in computer,which leads to it is replaced by 3D data gradually.Among all types of 3D data,the 3D point clouds are considered as an important data carrier in 3D model processing for their advantages on data acquisition and diverse geometric features.However,because of the lacking of orderliness and topology in point clouds,which makes it difficult to processing point clouds directly,the mesh reconstruction becomes the key of processing point clouds.The traditional methods of mesh reconstruction and feature extraction are often based on the differential invariant,which leads to large cost of calculation and high sensitivity to noise or thresholds,and a part of the methods are not sensitive to smooth features or tiny features.Thus,in order to tackle these barriers in current algorithms,a new mesh reconstruction based feature extraction method is proposed.The details are as follows:(1)Aiming at the problem that the current mesh reconstruction methods are sensitive to thresholds and noise,a new method named Adaptive Radius based-Alpha Shape(AR-Alpha Shape)is proposed.First,this algorithm calculates a radius for each point,while the radius is calculated based on corresponding point’s k nearest neighbors.Then,the proposed method generates local spheres based on the radius,which helps to connect the point with its neighbors.In order to improve the algorithm’s accuracy and robustness to the outliers,the algorithm introduces the constraint of adaptive upper bound,which prevent the outliers from connecting into the meshes.Then,based on the directed graph topology of the meshes,the algorithm removes meshes which contain one-way connected edges,which makes the reconstructed meshes become smoother.And as shown in the experiments,the AR-Alpha Shape is not sensitivity to thresholds and performances well on models which are large,irregular,complex and under different sampling density.(2)Aiming at the problem that the current feature extraction methods are sensitive to noisy and not sensitive to tiny or smooth features,a new method named Subgraph-based Local Binary Pattern(SGLBP)is proposed.First,the SGLBP algorithm extracts the topology and potential feature points with the AR-Alpha Shape on the point clouds.Then,the algorithm redefines the uniform patterns to separate non-feature points from potential feature points and removes the non-feature points iteratively until the SGLBP convergences.Finally,the SGLBP removes all the non-feature points and feature points are successfully retained.As shown in the experiments,the SGLBP is able to extract features on different kinds of point clouds.And the SGLBP is also able to deal with point clouds which are highly non-uniform sampled and contains more smooth features and tiny features.