A Simplification Method For Point Cloud Based On Feature Dimensionality Reduction And Fuzzy Cluster

With the continuous development of 3D sensor technology,the 3D scanning technology which can collect 3D point cloud data from object is getting more and more high precision and efficiency.At present,three-dimensional scanning technology has been widely used in the acquisition of three-dimensional spatial information by virtue of its non-contact,high-precision.However,there is a lot of redundancy in the massive cloud data obtained by scanning.In practical applications,the scale of original point cloud data obtained by 3D scanner is usually come to several hundred thousand or even millions,which increases the burden of point cloud data storage,transmission,computation and the difficulty of the subsequent processing.Therefore,a point cloud compression method is needed to compress the point cloud data as much as possible while preserving the feature information and guarantee the operation speed.In this paper,we focus on human body point cloud data compression,the human body point cloud model is composed of complex curved surface: the face and hand contain a lot of high curvature features information;while the abdomen and limbs trunk are mostly gentle surface,this complex point cloud model can better reflect the performance of compression algorithm.Currently,among common point cloud compression method,the method based on model topology is not applicable to human body model which contain complex surfaces;the method based on feature can compress the complex surface properly,but using feature like curvature would reduce the global feature information in the compressed model.Therefore,this paper introduces a new feature-PFH(Point Feature Histogram)and Fast Point Feature Histogram(FPFH).On the different geometrical positions(corner,plane)of point cloud model,FPFH has significant difference in distribution,so we call this feature "feature distinguish ability".In this paper,we propose an adaptive point cloud compression algorithm based on feature dimensionality reduction and fuzzy clustering,and obtain the following results:(1)Using the FPFH feature instead of the curvature feature,so that the compressed point cloud data can retain more global features.Then the PCA algorithm is used to reduce the dimensionality of FPFH to reduce the influence of "dimension disaster".We named dimension reduction characteristics as "PCA feature descriptor".It is proved by experiments that the PCA feature descriptor also has the "feature distinguishing ability",and the comparison experiment proves that we can get the best point cloud compression effect and surface reconstruction effect when select the PCA feature descriptor which can represent the original spatial sample information in the degree of 95% as sample data of cluster.(2)Achieve adaptive classification of point cloud data based on fuzzy C-means clustering algorithm: Firstly,we use PCA feature descriptor as the sample data of clustering algorithm,and its “feature distinguishing ability” can be used to classify the sample data into uniform PCA feature set(located on the corner of point cloud model,the feature point set)and inhomogeneous PCA feature set(located on the smooth surface of point cloud model,non-feature point set).As the PCA features correspond to the point cloud data,the set of point cloud can adaptively classified into two types: feature point set and non-feature point set.Finally,according to the established compression criterion,we give a big proportion compression to the non-feature point set to remove redundant points more,while giving a small proportion to the feature point set to retain feature information as much as possible.In this way,we can realize efficient point cloud compression.The traditional point cloud compression algorithm based on curvature feature need to set appropriate thresholds to classify different point sets in different curvature ranges.Our proposed compression method can achieve adaptive point cloud classification and compression.In addition,we can achieve different proportions of point cloud compression by varying the parameter ‘ratio'(the whole compression ratio of the point cloud);Meanwhile,we can change the retention ratio of the feature point set by adjusting the parameter ‘K'(Feature point set retention ratio: Non-feature point set retention ratio).Our compression algorithm can be much flexible by this way.(3)Propose an adaptive point cloud compression algorithm based on feature dimensionality reduction and fuzzy clustering.Our propose algorithm is evaluated and analyzed from three aspects: compression results at different compression rates,surface reconstruction effects after compression,and compression errors.Also,we compare algorithm effect with the traditional compression algorithm based on curvature characteristics.The experimental results show that the proposed point cloud compression algorithm has better compression performance under different compression ratios,and has certain accuracy in expressing shape.The feature information of the compressed point cloud model is well preserved and the contour information is still clearly visible.Comparing with the point cloud compression algorithm based on the curvature feature,the compression algorithm proposed in this paper can better preserve the details of the reconstruction model and have higher compression precision.In the end,we discuss the optimal aspects of this study and the future research direction.