CN114998456B - Three-dimensional point cloud attribute compression method based on local similarity - Google Patents

Three-dimensional point cloud attribute compression method based on local similarity Download PDF

Info

Publication number
CN114998456B
CN114998456B CN202210698973.1A CN202210698973A CN114998456B CN 114998456 B CN114998456 B CN 114998456B CN 202210698973 A CN202210698973 A CN 202210698973A CN 114998456 B CN114998456 B CN 114998456B
Authority
CN
China
Prior art keywords
point
attribute
point cloud
points
similarity
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202210698973.1A
Other languages
Chinese (zh)
Other versions
CN114998456A (en
Inventor
艾达
杨玉蓉
胥策
张晓阳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xian University of Posts and Telecommunications
Original Assignee
Xian University of Posts and Telecommunications
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Xian University of Posts and Telecommunications filed Critical Xian University of Posts and Telecommunications
Priority to CN202210698973.1A priority Critical patent/CN114998456B/en
Publication of CN114998456A publication Critical patent/CN114998456A/en
Application granted granted Critical
Publication of CN114998456B publication Critical patent/CN114998456B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T9/00—Image coding
    • G06T9/004—Predictors, e.g. intraframe, interframe coding
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G06T17/005—Tree description, e.g. octree, quadtree
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/761—Proximity, similarity or dissimilarity measures

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • Multimedia (AREA)
  • Computing Systems (AREA)
  • Geometry (AREA)
  • Health & Medical Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Computer Graphics (AREA)
  • Databases & Information Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Image Generation (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A three-dimensional point cloud attribute compression method based on local similarity comprises the steps of inputting point cloud data, sorting octree, generating predicted points, determining normal vector included angles between planes where the points are located, determining similarity of adjacent points, determining attribute predicted values and determining attribute predicted residual errors. By extracting the local data characteristics of the point cloud, constructing a local similarity calculation method based on Euclidean distance and normal angle, and selecting an actual attribute prediction value according to the maximum similarity value, the attribute information loss in the attribute prediction process can be reduced, and the prediction precision is improved. Through a comparative simulation experiment, the invention effectively improves the color attribute distortion degree in the compression process, has the characteristics of high compression efficiency, high reconstruction quality, easy realization and the like, is favorable for the storage and transmission of point clouds, and can be used for compression coding of colored point clouds.

Description

Three-dimensional point cloud attribute compression method based on local similarity
Technical Field
The invention belongs to the technical field of image processing, and particularly relates to a compression method of color attributes of three-dimensional point clouds.
Background
Three-dimensional point cloud data is one of the most representative three-dimensional data because of the advantages of high efficiency, high precision, easy acquisition and the like, and has been widely applied to various fields such as automatic driving, medical care, military national defense, cultural heritage, topographic exploration, industrial part production, virtual reality and augmented reality content creation and communication scenes. People can perceive and capture 3D scenes through a multi-view camera, a depth sensor, a laser radar scanner and the like, however, the huge data volume leads to limited storage and transmission of point clouds, so that the point cloud compression coding technology becomes a research hot spot in the field of computer vision.
In order to solve the application bottleneck of point cloud storage and transmission, a dynamic image expert group (Moving Picture Expert Group, MPEG) establishes an open point cloud compression standard system and issues a Geometry-based compression (G-PCC) test model. The attribute compression method of the point cloud mainly comprises three types: a transform-based attribute compression method, a prediction-based attribute compression method, and a mapping-based attribute compression method. The prediction-based method is suitable for compression of dense and sparse point clouds, and redundancy of attribute information can be effectively reduced.
The distortion degree of the compressed point cloud attribute determines the visual perception effect of human eyes on the three-dimensional point cloud, and the lower the distortion degree is, the more vivid the three-dimensional effect is observed by the human eyes. In recent years, attribute compression based on prediction has achieved numerous research results, but the rate-distortion performance of compression has yet to be improved.
Disclosure of Invention
The technical problem to be solved by the invention is to overcome the defects of the prior art, and provide a three-dimensional point cloud attribute compression method based on local similarity.
The technical scheme adopted for solving the technical problems is composed of the following steps:
(1) Inputting point cloud data
At point cloud P i ∈{P 1 ,P 2 ,...,P N In each point P i Each contains position information P i (x, y, z) and color attribute information P i (R, G, B), wherein N is a finite positive integer, x, y, z respectively represent three-dimensional coordinates, and R, G, B respectively represent red, green and blue.
(2) Octree ordering
Establishing a point cloud bounding box, and dividing a point cloud space into 2 layers w ×2 w ×2 w Each subcube has a side length of 2 determined by w ：
2 w ＝max(B x ,B y ,B z ) (1)
Wherein w is the division number, and the size of w is determined by the side length of the bounding box, B x ,B y ,B z And respectively representing the side lengths of bounding boxes, wherein in the dividing process, the subcubes containing the point cloud data are marked as 1, the subcubes without the point cloud data are marked as 0, and the occupied nodes marked as 1 can be continuously divided downwards until the number of points in each subcubes is at most m, and m is a preset value.
(3) Generating predicted points
Respectively determining M adjacent points P according to the formula (2) j J e {1,2,., M } is equal to the current point P i Is a Euclidean distance d (P) i ,P j )：
Figure BDA0003703218680000021
Wherein, (x) i ,y i ,z i ) For point P i Three-dimensional coordinates of (x) j ,y j ,z j ) For point P j Taking the distance P i The nearest K points are taken as K adjacent points, and the K adjacent points are taken as candidate predicted points, wherein K is smaller than M and K, M is a finite positive integer.
(4) Determining the angle of normal vector between planes of points
Determining the point P according to (3) i Plane and point P j The normal vector included angle theta (P) i ,P j )：
Figure BDA0003703218680000022
Wherein V is i And V j Respectively are points P i Sum point P j Is ∈ {1,2,., K }, is modulo.
(5) Determining similarity of adjacent points
Determining the adjacent point P according to the formula (4) j Relative to point P i Similarity S of (2) j ：
Figure BDA0003703218680000023
Wherein, alpha represents an angle weight factor, beta represents a distance weight factor, alpha+beta is 1, and alpha > beta.
(6) Determining attribute predictors
Determining a similarity maximum value among K adjacent points according to a formula (5):
S max ＝max(S j ) (5)
where j e {1,2,., K }, determining point P according to (6) i Attribute predictors of (2)
Figure BDA0003703218680000024
Figure BDA0003703218680000031
Wherein,,
Figure BDA0003703218680000032
for point P j Is>
Figure BDA0003703218680000033
Is the actual attribute value of the adjacent point with the maximum similarity, and
Figure BDA0003703218680000034
t is a similarity threshold.
(7) Determining attribute prediction residuals
Determining an attribute prediction residual ρ according to (7) i ：
Figure BDA0003703218680000035
And encoding the prediction residual error to realize attribute compression of the point cloud.
The step of generating the predicted point in the invention (3) is as follows:
respectively determining M adjacent points P according to the formula (2) j J e {1,2,., M } is equal to the current point P i Is a Euclidean distance d (P) i ,P j )：
Figure BDA0003703218680000036
Wherein, (x) i ,y i ,z i ) For point P i Three-dimensional coordinates of (x) j ,y j ,z j ) For point P j Taking the distance P i The nearest K points are taken as K adjacent points, and the K adjacent points are taken as candidate prediction points, wherein K is smaller than M, the K value is 3-5, and the M value is 20-30.
The step of determining the attribute predicted value in the invention (6) is as follows:
determining a similarity maximum value among K adjacent points according to a formula (5):
S max ＝max(S j ) (5)
where j e {1,2,., K }, determining point P according to (6) i Attribute predictors of (2)
Figure BDA00037032186800000312
Figure BDA0003703218680000037
Wherein,,
Figure BDA0003703218680000038
for point P j Is>
Figure BDA0003703218680000039
Is the actual attribute value of the adjacent point with the maximum similarity, and
Figure BDA00037032186800000310
t is the similarity threshold, and D is the average distance of the point cloud.
The step of determining the attribute prediction residual error in the invention (7) is as follows:
determining an attribute prediction residual according to (7)
Figure BDA00037032186800000311
Figure BDA0003703218680000041
The quantization parameters of the attribute residual comprise five quantization levels R1 to R5, and specific quantization values are 46, 40, 34, 28 and 22 respectively.
The octree ordering step (2) of the invention is:
establishing a point cloud bounding box, and dividing a point cloud space into 2 layers w ×2 w ×2 w Each subcube has a side length of 2 determined by w ：
2 w ＝max(B x ,B y ,B z ) (1)
Wherein w is the division number, and the size of w is determined by the side length of the bounding box, B x ,B y ,B z Representing the side lengths of bounding boxes, respectively, and in the partitioning process, subcubes containing point cloud data are marked as1, the subcubes without point cloud data are marked as 0, the occupied nodes marked as 1 can be continuously divided downwards until the number of points in each subcubes is at most m, m is a preset value, and the value of m is 1-100.
The beneficial effects of the invention are as follows:
the method is characterized in that the concave-convex change of the surface of the point cloud is fully utilized, the Euclidean distance and the normal angle between adjacent points are used as characteristic factors to construct a similarity model, two prediction modes are specified, and the final prediction mode is judged according to the similarity value. The simulation contrast experiment shows that the invention effectively improves the color attribute distortion degree in the compression process, has the characteristics of high compression efficiency, high reconstruction quality, easy realization and the like, is beneficial to the storage and transmission of point cloud, and can be used for compression coding of colored point cloud.
Drawings
Fig. 1 is a flow chart of the implementation of embodiment 1 of the present invention.
Fig. 2 is a schematic diagram of an input point cloud fande_00009_vox12 according to embodiment 1 of the present invention.
Fig. 3 is a schematic diagram of the fa ade_00009_vox12 point cloud reconstructed at the quantization level R1 according to embodiment 1 of the present invention.
Fig. 4 is a schematic diagram of the fa ade_00009_vox12 point cloud reconstructed at the quantization level R2 according to embodiment 1 of the present invention.
Fig. 5 is a schematic diagram of the fa ade_00009_vox12 point cloud reconstructed at the quantization level R3 according to embodiment 1 of the present invention.
Fig. 6 is a schematic diagram of the fa ade_00009_vox12 point cloud reconstructed at the quantization level R4 according to embodiment 1 of the present invention.
Fig. 7 is a schematic diagram of the reconstructed facade_00009vox12 point cloud at quantization level R5 according to embodiment 1 of the present invention.
Fig. 8 is a luminance rate distortion contrast curve of the method of embodiment 1 of the present invention compressing the facade_00009_vox12 with the G-PCC attribute prediction compression method.
Detailed description of the preferred embodiments
The present invention will be described in further detail with reference to the drawings and examples, but the present invention is not limited to the following embodiments.
Example 1
The three-dimensional point cloud attribute compression method based on local similarity of the embodiment is composed of the following steps (see fig. 1):
(1) Inputting point cloud data
At point cloud P i ∈{P 1 ,P 2 ,...,P N In each point P i Each contains position information P i (x, y, z) and color attribute information P i (R, G, B), wherein N is a finite positive integer, x, y, z respectively represent three-dimensional coordinates, and R, G, B respectively represent red, green and blue.
(2) Octree ordering
Establishing a point cloud bounding box, and dividing a point cloud space into 2 layers w ×2 w ×2 w Each subcube has a side length of 2 determined by w ：
2 w ＝max(B x ,B y ,B z ) (1)
Wherein w is the division number, and the size of w is determined by the side length of the bounding box, B x ,B y ,B z The side lengths of bounding boxes are respectively represented, in the dividing process, the subcubes containing point cloud data are marked as 1, the subcubes without point cloud data are marked as 0, the occupied nodes marked as 1 can be continuously divided downwards until the number of points in each subcubes is at most m, m is a preset value, the value of m is 1-100, and the value of m in the embodiment is 50.
(3) Generating predicted points
Respectively determining M adjacent points P according to the formula (2) j J e {1,2,., M } is equal to the current point P i Is a Euclidean distance d (P) i ,P j )：
Figure BDA0003703218680000051
Wherein, (x) i ,y i ,z i ) For point P i Three-dimensional coordinates of (x) j ,y j ,z j ) For point P j Taking the distance P i The nearest KThe point is taken as a K adjacent point, K adjacent points are taken as candidate predicted points, wherein K is smaller than M, K, M is a limited positive integer, the K value is 3-5, the M value is 20-30, the K value in the embodiment is 4, and the M value is 25.
(4) Determining the angle of normal vector between planes of points
Determining the point P according to (3) i Plane and point P j The normal vector included angle theta (P) i ,P j )：
Figure BDA0003703218680000052
Wherein V is i And V j Respectively are points P i Sum point P j Is ∈ {1,2,., K }, is modulo.
(5) Determining similarity of adjacent points
Determining the adjacent point P according to the formula (4) j Relative to point P i Similarity S of (2) j ：
Figure BDA0003703218680000061
Wherein, alpha represents an angle weight factor, beta represents a distance weight factor, alpha+beta is 1, alpha is larger than beta, alpha in the embodiment takes on a value of 0.825, and beta takes on a value of 0.175.
(6) Determining attribute predictors
Determining a similarity maximum value among K adjacent points according to a formula (5):
S max ＝max(S j ) (5)
where j e {1,2,., K }, determining point P according to (6) i Attribute predictors of (2)
Figure BDA0003703218680000062
Figure BDA0003703218680000063
Wherein,,
Figure BDA0003703218680000064
for point P j Is>
Figure BDA0003703218680000065
Is the actual attribute value of the adjacent point with the maximum similarity, and
Figure BDA0003703218680000066
d is the average distance of the point cloud, T is the similarity threshold, and the alpha value and the beta value are the same as those in the step (5).
(7) Determining attribute prediction residuals
Determining an attribute prediction residual according to (7)
Figure BDA0003703218680000067
Figure BDA0003703218680000068
And encoding the prediction residual error to realize attribute compression of the point cloud. The quantization parameters of the attribute residual comprise five quantization levels R1 to R5, and specific quantization values are 46, 40, 34, 28 and 22 respectively.
And (3) completing the three-dimensional point cloud attribute compression method based on the local similarity.
Example 2
The three-dimensional point cloud attribute compression method based on local similarity in the embodiment comprises the following steps:
(1) Inputting point cloud data
This step is the same as in example 1.
(2) Octree ordering
Establishing a point cloud bounding box, and dividing a point cloud space into 2 layers w ×2 w ×2 w Each subcube has a side length of 2 determined by w ：
2 w ＝max(B x ,B y ,B z ) (1)
Wherein w is the division number, and the size of w is determined by the side length of the bounding box, B x ,B y ,B z And respectively representing the side lengths of bounding boxes, wherein in the dividing process, the subcubes containing the point cloud data are marked as 1, the subcubes without the point cloud data are marked as 0, and the occupied nodes marked as 1 can be continuously divided downwards until the number of points in each subcubes is at most m, wherein m is a preset value, the value of m is 1-100, and the value of m in the embodiment is 1.
(3) Generating predicted points
Respectively determining M adjacent points P according to the formula (2) j J e {1,2,., M } is equal to the current point P i Is a Euclidean distance d (P) i ,P j )：
Figure BDA0003703218680000071
Wherein, (x) i ,y i ,z i ) For point P i Three-dimensional coordinates of (x) j ,y j ,z j ) For point P j Taking the distance P i The nearest K points are taken as K adjacent points, and the K adjacent points are taken as candidate prediction points, wherein K is smaller than M, K, M is a finite positive integer, the K value is 3-5, the M value is 20-30, the K value in the embodiment is 3, and the M value is 20.
(4) Determining the angle of normal vector between planes of points
This step is the same as in example 1.
(5) Determining similarity of adjacent points
Determining the adjacent point P according to the formula (4) j Relative to point P i Similarity S of (2) j ：
Figure BDA0003703218680000072
Wherein, alpha represents an angle weight factor, beta represents a distance weight factor, alpha+beta is 1, alpha is larger than beta, alpha in the embodiment takes on the value of 0.694, and beta takes on the value of 0.306.
The other steps were the same as in example 1. And (3) completing the three-dimensional point cloud attribute compression method based on the local similarity.
Example 3
The three-dimensional point cloud attribute compression method based on local similarity in the embodiment comprises the following steps:
(1) Inputting point cloud data
This step is the same as in example 1.
(2) Octree ordering
Establishing a point cloud bounding box, and dividing a point cloud space into 2 layers w ×2 w ×2 w Each subcube has a side length of 2 determined by w ：
2 w ＝max(B x ,B y ,B z ) (1)
Wherein w is the division number, and the size of w is determined by the side length of the bounding box, B x ,B y ,B z And respectively representing the side lengths of bounding boxes, wherein in the dividing process, the subcubes containing the point cloud data are marked as 1, the subcubes without the point cloud data are marked as 0, and the occupied nodes marked as 1 can be continuously divided downwards until the number of points in each subcubes is at most m, wherein m is a preset value, the value of m is 1-100, and the value of m in the embodiment is 100.
(3) Generating predicted points
Respectively determining M adjacent points P according to the formula (2) j J e {1,2,., M } is equal to the current point P i Is a Euclidean distance d (P) i ,P j )：
Figure BDA0003703218680000081
Wherein, (x) i ,y i ,z i ) For point P i Three-dimensional coordinates of (x) j ,y j ,z j ) For point P j Taking the distance P i The nearest K points are taken as K adjacent points, and the K adjacent points are taken as candidatesWherein K is less than M, K, M is a finite positive integer, the K value is 3-5, the M value is 20-30, the K value in this embodiment is 5, and the M value is 30.
(4) Determining the angle of normal vector between planes of points
This step is the same as in example 1.
(5) Determining similarity of adjacent points
Determining the adjacent point P according to the formula (4) j Relative to point P i Similarity S of (2) j ：
Figure BDA0003703218680000091
Wherein, alpha represents an angle weight factor, beta represents a distance weight factor, alpha+beta is 1, alpha is larger than beta, alpha in the embodiment takes on the value of 0.547, and beta takes on the value of 0.453.
The other steps were the same as in example 1. And (3) completing the three-dimensional point cloud attribute compression method based on the local similarity.
In order to prove the beneficial effects of the invention, the inventor adopts the three-dimensional point cloud compression method based on local similarity and the G-PCC attribute compression method based on prediction (hereinafter referred to as G-PCC Pred) of the embodiment 1 of the invention to carry out a comparison experiment, and the experimental conditions are as follows:
fig. 2 is a standard point cloud face00009_vox12 provided by MPEG, which has rich texture variation, containing 1596085 points, each including position information (x, y, z) and color attribute information (R, G, B). The attribute compression test was performed as in example 1, with objective evaluation and subjective visual reconstruction. Y is the brightness index of the color, and the evaluation index comprises: the peak signal-to-noise ratio (hereinafter abbreviated as Y-PSNR), bit rate (hereinafter abbreviated as Bitrate), BD-PSNR of Y are shown in table 1 and fig. 8, and the subjective evaluation results are shown in fig. 3 to 7.
TABLE 1 Point cloud Facade_00009_vox12 compression results
Figure BDA0003703218680000092
Table 1 shows compression performance indexes of compressing the point cloud fande_00009_vox12 under different quantization parameters, and under different quantization parameters, the method of the present invention effectively improves the PSNR value of the reconstructed point cloud Y channel, and fig. 8 is a luminance attribute compression rate distortion graph of example 1, compared with the G-PCC Pred compression method, BD-PSNR is 0.55dB. The compression effect of five quantization levels of R1-R5 is shown in fig. 3-7, and as the quantization levels are raised, the reconstruction point cloud is closer to the original point cloud, and more color information is reserved. The method effectively reserves the color information of the point cloud, improves the compression quality and has better rate distortion performance.
In summary, experimental results show that compared with the G-PCC Pred compression method, the method provided by the invention effectively improves the color attribute distortion degree in the compression process, has the characteristics of high compression efficiency, high reconstruction quality, easiness in realization and the like, is beneficial to the storage and transmission of point clouds, and can be used for compression coding of colored point clouds.

Claims (5)

1.一种基于局部相似度的三维点云属性压缩方法，其特征由下述步骤组成：1. A method for compressing three-dimensional point cloud attributes based on local similarity, characterized by the following steps: (1)输入点云数据(1) Input point cloud data 在点云Pi∈{P1,P2,...,PN}中，每个点Pi均包含位置信息Pi(x,y,z)及颜色属性信息Pi(R,G,B)，其中，N为有限的正整数，x、y、z分别代表三维坐标，R、G、B分别表示红色、绿色、蓝色；In the point cloud P i ∈{P 1 ,P 2 ,...,P N }, each point P i contains position information P i (x,y,z) and color attribute information P i (R,G , B), wherein, N is a finite positive integer, x, y, z represent three-dimensional coordinates, R, G, B represent red, green, blue respectively; (2)八叉树排序(2) Octree sorting 建立点云包围盒，将点云空间逐层划分成2w×2w×2w个子立方体，按下式确定每个子立方体边长2w：Establish the point cloud bounding box, divide the point cloud space into 2 w × 2 w × 2 w sub-cubes layer by layer, and determine the side length 2 w of each sub-cube according to the following formula: 2w＝max(Bx,By,Bz) (1)2 w ＝max(B x ,B y ,B z ) (1) 其中，w是划分次数，w的大小由包围盒的边长决定，Bx,By,Bz分别表示包围盒的边长，在划分过程中，包含点云数据的子立方体标记为1，无点云数据的子立方体标记为0，被标记为1的占用节点可继续向下划分，直至每个子立方体内的点数最多为m，m为预设的值；Among them, w is the number of divisions, and the size of w is determined by the side length of the bounding box. B x , By y , and B z represent the side lengths of the bounding box respectively. During the division process, the sub-cube containing point cloud data is marked as 1, The sub-cube without point cloud data is marked as 0, and the occupied nodes marked as 1 can continue to be divided down until the number of points in each sub-cube is at most m, and m is a preset value; (3)生成预测点(3) Generate prediction points 按式(2)分别确定M个近邻点Pj,j∈{1,2,...,M}与当前点Pi的欧式距离d(Pi,Pj)：Determine the Euclidean distance d(P i ,P j ) between M neighboring points P j ,j∈{1,2,...,M} and the current point P i according to formula (2):
Figure QLYQS_1
Figure QLYQS_1
 其中，(xi,yi,zi)为点Pi的三维坐标，(xj,yj,zj)为点Pj的三维坐标，取距离Pi最近的K个点作为其K近邻点，将K个近邻点作为候选的预测点，其中，K＜M，K、M为有限的正整数；Among them, (x i , y i , z i ) are the three-dimensional coordinates of point P i , (x j , y j , z j ) are the three-dimensional coordinates of point P j , and the K points closest to P i are taken as its K Neighbor points, using K neighbor points as candidate prediction points, where K<M, K and M are finite positive integers; (4)确定点所在平面间的法向量夹角(4) Determine the normal vector angle between the planes where the points are located 按式(3)确定点Pi所在平面与点Pj所在平面的法向量夹角θ(Pi,Pj)：Determine the normal vector angle θ(P i , P j ) between the plane where point P i is located and the plane where point P j is located according to formula (3):
Figure QLYQS_2
Figure QLYQS_2
 其中，Vi和Vj分别为点Pi和点Pj的法向量，j∈{1,2,...,K}，||·||表示模；Among them, V i and V j are the normal vectors of point P i and point P j respectively, j∈{1,2,...,K}, |||| (5)确定近邻点的相似度(5) Determine the similarity of neighboring points 按式(4)确定近邻点Pj相对于点Pi的相似度Sj：Determine the similarity S j of the neighboring point P j relative to the point P i according to formula (4):
Figure QLYQS_3
Figure QLYQS_3
 其中，α表示角度权重因子，β表示距离权重因子，α+β为1，且α＞β；Among them, α represents the angle weight factor, β represents the distance weight factor, α+β is 1, and α>β; (6)确定属性预测值(6) Determine the attribute prediction value 按式(5)确定K个近邻点中的相似度最大值：According to formula (5), determine the maximum similarity among the K neighbor points: Smax＝max(Sj) (5)S max = max(S j ) (5) 其中，j∈{1,2,...,K}，按式(6)确定点Pi的属性预测值
Figure QLYQS_4
Among them, j∈{1,2,...,K}, according to formula (6) to determine the attribute prediction value of point P i
Figure QLYQS_4
Figure QLYQS_5
Figure QLYQS_5
 其中，
Figure QLYQS_6
为点Pj的实际属性值，/>
Figure QLYQS_7
为相似度最大的近邻点的实际属性值，且
Figure QLYQS_8
T为相似度阈值；in,
Figure QLYQS_6
is the actual attribute value of point P j , />
Figure QLYQS_7
is the actual attribute value of the neighbor point with the largest similarity, and
Figure QLYQS_8
T is the similarity threshold; (7)确定属性预测残差(7) Determine the attribute prediction residual 按式(7)确定属性预测残差
Figure QLYQS_9
Determine the attribute prediction residual according to formula (7)
Figure QLYQS_9
Figure QLYQS_10
Figure QLYQS_10
 对预测残差进行编码实现点云的属性压缩。Encoding the prediction residuals enables attribute compression of point clouds. 2.根据权利要求1所述的基于局部相似度的三维点云属性压缩方法，其特征在于所述的(3)生成预测点步骤为：2. the three-dimensional point cloud attribute compression method based on local similarity according to claim 1, is characterized in that described (3) generation prediction point step is: 按式(2)分别确定M个近邻点Pj,j∈{1,2,...,M}与当前点Pi的欧式距离d(Pi,Pj)：Determine the Euclidean distance d(P i ,P j ) between M neighboring points P j ,j∈{1,2,...,M} and the current point P i according to formula (2):
Figure QLYQS_11
Figure QLYQS_11
 其中，(xi,yi,zi)为点Pi的三维坐标，(xj,yj,zj)为点Pj的三维坐标，取距离Pi最近的K个点作为其K近邻点，将K个近邻点作为候选的预测点，其中，K＜M，K值取为3～5，M值取为20～30。Among them, (x i , y i , z i ) are the three-dimensional coordinates of point P i , (x j , y j , z j ) are the three-dimensional coordinates of point P j , and the K points closest to P i are taken as its K Neighboring points, K neighboring points are used as candidate prediction points, wherein, K<M, the value of K is 3-5, and the value of M is 20-30. 3.根据权利要求1所述的基于局部相似度的三维点云属性压缩方法，其特征在于所述的(6)确定属性预测值步骤为：3. the three-dimensional point cloud attribute compression method based on local similarity according to claim 1, is characterized in that described (6) determines attribute prediction value step is: 按式(5)确定K个近邻点中的相似度最大值：According to formula (5), determine the maximum similarity among the K neighbor points: Smax＝max(Sj) (5)S max = max(S j ) (5) 其中，j∈{1,2,...,K}，按式(6)确定点Pi的属性预测值
Figure QLYQS_12
Among them, j∈{1,2,...,K}, according to formula (6) to determine the attribute prediction value of point P i
Figure QLYQS_12
Figure QLYQS_13
Figure QLYQS_13
Figure QLYQS_14
Figure QLYQS_14
 其中，
Figure QLYQS_15
为点Pj的实际属性值，/>
Figure QLYQS_16
为相似度最大的近邻点的实际属性值，且
Figure QLYQS_17
T为相似度阈值，D为点云的平均距离。in,
Figure QLYQS_15
is the actual attribute value of point P j , />
Figure QLYQS_16
is the actual attribute value of the neighbor point with the largest similarity, and
Figure QLYQS_17
T is the similarity threshold, and D is the average distance of the point cloud. 4.根据权利要求1所述的基于局部相似度的三维点云属性压缩方法，其特征在于所述的(7)确定属性预测残差步骤为：4. the three-dimensional point cloud attribute compression method based on local similarity according to claim 1, is characterized in that described (7) determines attribute prediction residual error step is: 按式(7)确定属性预测残差
Figure QLYQS_18
Determine the attribute prediction residual according to formula (7)
Figure QLYQS_18
Figure QLYQS_19
Figure QLYQS_19
 属性残差的量化参数包括R1～R5五个量化等级，具体量化值分别为46、40、34、28、22。The quantization parameters of attribute residuals include five quantization levels R1 to R5, and the specific quantization values are 46, 40, 34, 28, and 22, respectively. 5.根据权利要求1所述的基于局部相似度的三维点云属性压缩方法，其特征在于所述的(2)八叉树序步骤为：5. the three-dimensional point cloud attribute compression method based on local similarity according to claim 1, is characterized in that described (2) octree order step is: 建立点云包围盒，将点云空间逐层划分成2w×2w×2w个子立方体，按下式确定每个子立方体边长2w：Establish the point cloud bounding box, divide the point cloud space into 2 w × 2 w × 2 w sub-cubes layer by layer, and determine the side length 2 w of each sub-cube according to the following formula: 2w＝max(Bx,By,Bz) (1)2 w ＝max(B x ,B y ,B z ) (1) 其中，w是划分次数，w的大小由包围盒的边长决定，Bx,By,Bz分别表示包围盒的边长，在划分过程中，包含点云数据的子立方体标记为1，无点云数据的子立方体标记为0，被标记为1的占用节点可继续向下划分，直至每个子立方体内的点数最多为m，m为预设的值，m取值为1～100。Among them, w is the number of divisions, and the size of w is determined by the side length of the bounding box. B x , By y , and B z represent the side lengths of the bounding box respectively. During the division process, the sub-cube containing point cloud data is marked as 1, The sub-cube without point cloud data is marked as 0, and the occupied nodes marked as 1 can continue to be divided down until the number of points in each sub-cube is at most m, m is a preset value, and m ranges from 1 to 100.
CN202210698973.1A 2022-06-20 2022-06-20 Three-dimensional point cloud attribute compression method based on local similarity Active CN114998456B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210698973.1A CN114998456B (en) 2022-06-20 2022-06-20 Three-dimensional point cloud attribute compression method based on local similarity

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210698973.1A CN114998456B (en) 2022-06-20 2022-06-20 Three-dimensional point cloud attribute compression method based on local similarity

Publications (2)

Publication Number Publication Date
CN114998456A CN114998456A (en) 2022-09-02
CN114998456B true CN114998456B (en) 2023-06-30

Family

ID=83034407

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210698973.1A Active CN114998456B (en) 2022-06-20 2022-06-20 Three-dimensional point cloud attribute compression method based on local similarity

Country Status (1)

Country Link
CN (1) CN114998456B (en)

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111866518A (en) * 2020-07-29 2020-10-30 西安邮电大学 Adaptive 3D point cloud compression method based on feature extraction
CN113096200A (en) * 2021-03-24 2021-07-09 中山大学 Point cloud attribute compression method, device, equipment and medium based on normal vector
CN113424547A (en) * 2019-03-20 2021-09-21 腾讯美国有限责任公司 Techniques and apparatus for weighted median prediction for point cloud attribute encoding and decoding
WO2021246843A1 (en) * 2020-06-05 2021-12-09 엘지전자 주식회사 Point cloud data transmission device, point cloud data transmission method, point cloud data reception device, and point cloud data reception method
CN113868476A (en) * 2021-09-28 2021-12-31 西安热工研究院有限公司 Octree point cloud preprocessing method based on local density
CN114066953A (en) * 2021-11-22 2022-02-18 南京大学 A Deformable Registration Method for 3D Multimodal Images for Rigid Targets
CN114331989A (en) * 2021-12-16 2022-04-12 重庆邮电大学 Full-reference 3D point cloud quality assessment method based on point feature histogram geodetic distance
CN114549870A (en) * 2022-02-17 2022-05-27 山东大学 Dynamic point cloud rapid global motion matching method and system based on centroid point search

Patent Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113424547A (en) * 2019-03-20 2021-09-21 腾讯美国有限责任公司 Techniques and apparatus for weighted median prediction for point cloud attribute encoding and decoding
WO2021246843A1 (en) * 2020-06-05 2021-12-09 엘지전자 주식회사 Point cloud data transmission device, point cloud data transmission method, point cloud data reception device, and point cloud data reception method
CN111866518A (en) * 2020-07-29 2020-10-30 西安邮电大学 Adaptive 3D point cloud compression method based on feature extraction
CN113096200A (en) * 2021-03-24 2021-07-09 中山大学 Point cloud attribute compression method, device, equipment and medium based on normal vector
CN113868476A (en) * 2021-09-28 2021-12-31 西安热工研究院有限公司 Octree point cloud preprocessing method based on local density
CN114066953A (en) * 2021-11-22 2022-02-18 南京大学 A Deformable Registration Method for 3D Multimodal Images for Rigid Targets
CN114331989A (en) * 2021-12-16 2022-04-12 重庆邮电大学 Full-reference 3D point cloud quality assessment method based on point feature histogram geodetic distance
CN114549870A (en) * 2022-02-17 2022-05-27 山东大学 Dynamic point cloud rapid global motion matching method and system based on centroid point search

Non-Patent Citations (9)

* Cited by examiner, † Cited by third party
Title
3D彩色点云高效压缩算法及质量评价研究;张习民;中国博士学位论文全文数据库信息科技辑;1-62 *
An Adaptive Feature-based Quantization Algorithm for Point Cloud Compression;Da Ai等;2021 Picture Coding Symposium (PCS);1-5 *
Inferring Point Cloud Quality via Graph Similarity;Qi Yang等;IEEE Transactions on Pattern Analysis and Machine Intelligence;3015 - 3029 *
Towards a Point Cloud Structural Similarity Metric;Evangelos Alexiou等;2020 IEEE International Conference on Multimedia & Expo Workshops (ICMEW);1-6 *
一种散乱点云的均匀精简算法;李仁忠;杨曼;刘阳阳;张缓缓;;光学学报(07);97-105 *
向量相似度可复原三维点云压缩算法;张旭康;牛保宁;张锦文;;计算机科学与探索(04);121-132 *
基于FPFH特征的三维场景重建;艾达;王苗;倪国斌;;计算机测量与控制(07);240-244 *
基于Kinect的三维重建技术综述;艾达;倪国斌;王苗;杨珍;;传感器与微系统(08);6-11 *
基于形状分析的三维点云模型压缩;何辰;中国博士学位论文全文数据库信息科技辑;1-66 *

Also Published As

Publication number Publication date
CN114998456A (en) 2022-09-02

Similar Documents

Publication Publication Date Title
CN114424542B (en) Video-based point cloud compression with non-canonical smoothing
Schwarz et al. Emerging MPEG standards for point cloud compression
CN111837392B (en) Processing point clouds
US11509933B2 (en) Method, an apparatus and a computer program product for volumetric video
CN109257604A (en) A kind of color attribute coding method based on TMC3 point cloud encoder
CN104756489B (en) A kind of virtual visual point synthesizing method and system
CN103248909B (en) Method and system of converting monocular video into stereoscopic video
CN111433821A (en) Method and apparatus for reconstructing a point cloud representing a 3D object
CN111971969B (en) Method and apparatus for encoding the geometry of a point cloud
JP7477528B2 (en) Point Cloud Processing
CN113518226A (en) G-PCC point cloud coding improvement method based on ground segmentation
WO2019199726A1 (en) A method for encoding depth values of a set of 3d points once orthogonally projected into at least one image region of a projection plane
EP3777182A1 (en) A method and apparatus for encoding/decoding a point cloud representing a 3d object
US20210112277A1 (en) A method and apparatus for encoding/decoding a point cloud representing a 3d object
CN106162198A (en) 3 D video depth map encoding based on the most homogeneous piece of segmentation and coding/decoding method
Wei et al. Weighted attribute prediction based on morton code for point cloud compression
Faramarzi et al. Mesh coding extensions to mpeg-i v-pcc
CN114998456B (en) Three-dimensional point cloud attribute compression method based on local similarity
CN107592538B (en) A method of reducing stereoscopic video depth map encoder complexity
Tohidi et al. Dynamic point cloud compression with cross-sectional approach
Zhang et al. Point cloud attribute compression via clustering and intra prediction
EP4133719A1 (en) A method, an apparatus and a computer program product for volumetric video coding
WO2020237759A1 (en) System and method for effectively representing and coding image
CN104394399B (en) Three limit filtering methods of deep video coding
US20230171427A1 (en) Method, An Apparatus and a Computer Program Product for Video Encoding and Video Decoding

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant