CN110363822A - A kind of 3D point cloud compression method - Google Patents
A kind of 3D point cloud compression method Download PDFInfo
- Publication number
- CN110363822A CN110363822A CN201810322696.8A CN201810322696A CN110363822A CN 110363822 A CN110363822 A CN 110363822A CN 201810322696 A CN201810322696 A CN 201810322696A CN 110363822 A CN110363822 A CN 110363822A
- Authority
- CN
- China
- Prior art keywords
- point
- individual
- class
- cloud
- population
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T19/00—Manipulating 3D models or images for computer graphics
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Graphics (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Processing (AREA)
Abstract
The present invention provides a kind of 3D point cloud compaction coding method, this method includes a kind of hierarchical clustering algorithm for cloud, and for the point put in cloud to be divided into the class with different attribute, the point in same class, each attribute is all similar；Comprising a kind of optimal mapping algorithm that cloud is mapped as to two dimensional image, to compress point cloud data using efficient image encoding method.The two dimensional image that cloud is mapped to rule is greatly improved the performance and efficiency of point cloud compression algorithm to compress point cloud data using Image Compression by the present invention.
Description
Technical field
The present invention relates to 3D media compression coding fields, more particularly, to a kind of 3D point cloud compression method.
Background technique
With the fast development of three dimensional data collection equipment, three-dimension object is described with 3D point cloud, reproduction three-dimensional scenic becomes
It is more and more convenient and efficient.3D point cloud is a kind of new data format, for recording and indicating the surface information of three-dimension object.Point cloud
Data are the set of series of points in space, these points include three-dimensional coordinate and one or more attribute informations, such as color, normal direction
Amount, reflected intensity etc..Compared with other three-dimensional data formats, point cloud, which has, obtains the advantages such as convenient, processing is simple, therefore wide
It is applied to various emerging fields, such as augmented reality (AR), automatic Pilot and 3D printing generally.But it is passed with image, video etc.
System data mode is compared, and the data volume for putting cloud is very big.On the one hand, the dimension of point cloud data is higher, commonly believes comprising color
The point cloud of breath is sextuple data, if other attributes are added, dimension is higher.On the other hand, the points that point cloud is included are also very big.
For example, building immersion experience to really describe three-dimension object, applied to the point cloud in AR, points are usually million
Magnitude, it is even higher.Such huge data volume all brings great challenge to storage, processing and transmission, meanwhile, also limit
Some clouds have been made to the application in the higher field of requirement of real-time.Therefore, point cloud data feature and internal information are being ensured
On the basis of, amount of compressed data is the inevitable choice for being further processed point cloud data as much as possible.
The point cloud compression technology majority being widely used at present is realized based on Octree spatial decomposition.Utilize Octree
Structure decomposes the three-dimensional space where a cloud, and the center approximation in the space representated by each child node replaces it to be wrapped
Position containing point, the geometry of the point cloud after approximation can be calculated according to octree structure and corresponding bounding box information.
Therefore, point cloud compression can be realized by carrying out coding to octree structure.This method solves the problems, such as to a certain extent,
Reduce data volume.But this method inevitably introduces the distortion of geometry, therefore is not suitable for 3D printing, text
Object reparation etc. is higher to required precision or the application of lossless compression.In addition, the compression ratio of this method is also far from enough, need to be mentioned
It rises.
In addition, technology is highly developed after decades of development for compression of images, many algorithms are simple and efficient, if fortune
It uses in point cloud compression, the performance and efficiency of point cloud compression algorithm will be greatly improved.
Summary of the invention
For the defects in the prior art, the 3D point cloud based on hierarchical cluster and mapping that the object of the present invention is to provide a kind of
Compression method, so that it is low to solve compression ratio in existing method, it is difficult to the problems such as realizing lossless compression.
The present invention provides a kind of 3D point cloud compaction coding method, this method includes a kind of hierarchical cluster calculation for cloud
Method, for the point put in cloud to be divided into the class with different attribute, the point in same class, each attribute is all similar；Include one
Point cloud (class after division) is mapped as the optimal mapping algorithm of two dimensional image by kind, to utilize efficient image encoding method pressure
Contracting point cloud data, specifically,
A kind of 3D point cloud compaction coding method, which comprises the steps of:
Step S1: by the point divide into several classes in cloud, and make the point either space coordinate or face in each class
Color attribute is all very close；
Step S2: and then compressed encoding is carried out to each class respectively again.
In above-mentioned technical proposal, the step S1 is sought by the way that central point is constantly moved to the higher region of probability density
It looks for and belongs to of a sort data point, specifically comprise the following steps:
Step S101: input original point cloud data randomly selects a point as initial center in not labeled point
Point mn；
Step S102: with mnCentered on point creation one class Ci；
Step S103: search in all input data points with mnThe distance between be less than radius r all neighborhood points, by this
Labeled as having accessed, each point can be accessed repeatedly a little points, and update each point by class CiThe number of access；
Step S104: m is calculatednWith the weighted mass center m of all neighborhood points searched outn+1, and central point is updated to mn+1；
Step S105: if mn+1And mnThe distance between be greater than the threshold value of setting, then return step (3)；Otherwise, iteration
Convergence, into next step；
Step S106: if the central point after convergence is less than the threshold value of setting at a distance from the central point of existing class,
It is merged with already present class；
Step S107: above step is repeated until all the points are all labeled；
Step S108: the number that each point is accessed by each class, class of the class for taking access times most as the point are read；
Step S109: the point cloud data after output category.
In above-mentioned technical proposal, the data that cloud is included are mapped to the two dimensional image of rule by step S2, are then utilized
Image Coding Algorithms carry out compressed encoding, set a class C containing m pointj={ x1,x2,…,xm, it finds and carries out zigzag
The C ' that puts in order when type mapsj={ x(1),x(2),…,x(m), made under the sequence using genetic algorithmValue it is minimum, a point in each gene representation class, individual expression class comprising m gene
A kind of arrangement mode, multiple individuals constitute a population, and defining fitness individual in population is f=
The sum of the distance of i.e. all consecutive points is smaller, and individual fitness is higher, to there is more maximum probability to survive.
In above-mentioned technical proposal, the genetic algorithm is mainly comprised the steps that
Step 201: a class after input cluster segmentation, the point in random alignment class generate n individual, constitute initially
Population；
Step S202: the fitness of each individual is calculated
Step S203: the individual that all fitness in population are less than given threshold s is eliminated, and will be under the individual addition of survival
Generation population；
Step S204: two individuals are randomly choosed from the individual of survival with Probability p1Hybridized, generate new individual,
And next-generation population is added, wherein hybridizing method are as follows: the gene grafting of one section of random-length is intercepted from the leading portion of an individual
The leading portion individual to second, then remove and do not repeated mutually with the interception duplicate gene in part, the gene of the newly-generated individual of guarantee；
Step S205: with Probability p2Gene mutation is carried out to newly-generated individual.Gene mutation method are as follows: in individual two
A gene carries out place-exchange；
Step 206: step S204 and S205 are repeated, until the individual of number and initial population individual in population of new generation
Number is equal；
Step S207: repeat step S202 to S206 until in population the highest individual of fitness meet predetermined condition, it is defeated
The highest individual of fitness out；
Step 208: output meets the point sequence of optimal mapping.
Compared with prior art, the present invention have it is following the utility model has the advantages that
Cloud is mapped to the two dimensional image of rule by the present invention, so that point cloud data is compressed using Image Compression,
Greatly improve the performance and efficiency of point cloud compression algorithm.
Detailed description of the invention
Upon reading the detailed description of non-limiting embodiments with reference to the following drawings, other feature of the invention,
Objects and advantages will become more apparent upon:
Fig. 1 is the schematic diagram of compression method proposed by the invention；
Fig. 2 is the schematic diagram of hierarchical cluster partitioning algorithm, wherein (a) is input point cloud, (b) after for first layer segmentation
Point cloud is (c) the point cloud after second layer segmentation, is (d) two classes in (b) with (e), is (f) (e) after the second layer is divided
As a result；
Fig. 3 is mapping schematic diagram, wherein (a) is the signal of zigzag mode, (b) is Mapping of data points mode.
Specific embodiment
The present invention is described in detail combined with specific embodiments below.Following embodiment will be helpful to the technology of this field
Personnel further understand the present invention, but the invention is not limited in any way.It should be pointed out that the ordinary skill of this field
For personnel, without departing from the inventive concept of the premise, several changes and improvements can also be made.These belong to the present invention
Protection scope.
A kind of 3D point cloud compaction coding method of the invention, which comprises the steps of:
Step S1: by the point divide into several classes in cloud, and make the point either space coordinate or face in each class
Color attribute is all very close；
Step S2: and then compressed encoding is carried out to each class respectively again.
Wherein, divided in step S1 using hierarchical cluster.
Point cloud generally comprises ten hundreds of or even million meters points, and the attribute differences such as the spatial position of these points and color
Different larger, directly compression calculation amount is larger and compression performance is poor.If by the point divide into several classes in cloud, and making each
Point either space coordinate or color attribute in class is all very close, then compresses respectively to each class again, can be big
It is big to improve point cloud compression performance.Since cloud has sparsity and randomness, the dot density of each position exists in three-dimensional space
Difference, therefore the present invention uses the mean shift algorithm to match with this characteristic as clustering algorithm.The algorithm is a kind of
Gradient ascent algorithm, found by the way that central point (mean) is constantly moved to the higher region of probability density belong to it is of a sort
Data point.Specific step is as follows for Mean shift algorithm:
Algorithm input are as follows: original point cloud data
Algorithm output are as follows: sorted point cloud data
(1) point is randomly selected as initial center point m in not labeled pointn；
(2) with mnCentered on point creation one class Ci；
(3) search in all input data points with mnThe distance between be less than radius r all neighborhood points, by these point mark
It is denoted as and has accessed, each point can be accessed repeatedly, and update each point by class CiThe number of access；
(4) m is calculatednWith the weighted mass center m of all neighborhood points searched outn+1, and central point is updated to mn+1
(5) if mn+1And mnThe distance between be greater than the threshold value of setting, then return step (3)；Otherwise, iteration convergence, into
Enter in next step；
(6) if convergence after central point with the central point of existing class at a distance from be less than set threshold value, by its with
Already present class merges；
(7) above step is repeated until all the points are all labeled；
(8) number that each point is accessed by each class, class of the class for taking access times most as the point are read.
Mean shift is a kind of unsupervised segmentation algorithm, can be classified to each point in input point cloud, and nothing
Cluster number need to be specified.But traditional mean shift algorithm is used on a cloud that there is also certain problems.For most often
It include the point cloud (X={ x of geological information and colouring informationi,y,zi,ri,gi,bi, i=1 ..., n), if by mean
Shift algorithm expands to three-dimensional space and clusters to geological information (XYZ), it will in the class after leading to segmentation, Ge Gedian
Between spatial position it is close, but color and dissimilar.If mean shift algorithm is expanded to sextuple space (XYZRGB), together
When geological information and colouring information are clustered, it will cause very poor in boundary classification results.For this purpose, the present invention proposes one
Hierarchical cluster structure is planted to solve the problems, such as that traditional mean shift algorithm generates in cloud cluster.
Commonly to include the point cloud (X={ x of geological information and colouring informationi,y,zi,ri,gi,bi, i=1 ..., n) be
Example, since point cloud data to be compressed only includes geological information and colouring information, our hierarchical cluster partitioning algorithm packet
Containing double-layer structure, mean shift cluster is carried out in two different spaces.First layer is to cluster in color space to a cloud.
When field is searched for, global search is carried out to all the points according to rgb coordinate value in entirely point cloud.For original shown in Fig. 2 (a)
Initial point cloud, total result such as Fig. 2 (b) after first layer clusters are shown.In Fig. 2, (d) and (e) is two classes in (b).It can be with
Find out, in original point cloud, two different colors of pattern is accurately assigned to two different classes on skirt.But only
First layer cluster there is also some problems.For example, since skirt is similar with shoes color, they are divided in Fig. 2 (e)
Into same class.Obviously, the point represented by them, farther out, geometric coordinate value difference is not larger, Yao Tigao for distance in geometric space
Compression ratio, they should be assigned in different classes.Therefore, we carry out second layer cluster, are sat in geometric space according to xyz
Scale value carries out further cluster segmentation to each class that first layer generates.Skirt and shoes as shown in Fig. 2 (f), in Fig. 2 (e)
It is separated after second layer cluster, and is divided into smaller piece, to reduce the calculation amount of subsequent processes.After second layer cluster
Result such as Fig. 2 (c) shown in.Not only space length is close for the point for including in each class after hierarchical cluster segmentation, but also color phase
Seemingly.
Hierarchical cluster dividing method proposed by the invention, it is not limited to two layers of cluster, but had by be split cloud
The attribute that body includes determines.For example, for the point cloud data comprising three attribute such as geological information, colouring information and normal vector,
It then should include three layers of cluster segmentation.
After decades of development, technology is quite mature, and many algorithms are simple and efficient, if be applied to for compression of images
In point cloud compression, the performance and efficiency of point cloud compression algorithm will be greatly improved.Therefore, the present invention reflects the data that a cloud is included
The two dimensional image of rule is penetrated into, then carries out compressed encoding using Image Coding Algorithms.
It wants matching image to compress, improves the compression ratio of algorithm, point cloud data is when to Planar Mapping, it is ensured that these points exist
The value of the similitude of two-dimensional space, i.e., the value Ying Yuqi neighborhood point of each point is as close as possible.It is calculated in this way using compression of images
When method, can guarantee can obtain preferable compression performance.To be further simplified problem, the complexity of algorithm is reduced, the present invention adopts
It is mapped with zigzag type shown in Fig. 3.By zigzag mode map, as long as guaranteeing similar between the point of front and back in one-dimensional sequence
Property, that is, it can guarantee the similitude at two dimensional image midpoint after mapping.Pattern type similar compared to direct construction two dimension, Zigzag type reflect
The problem of enormously simplifying is penetrated, the complexity of algorithm is reduced.
Point cloud data has scrambling, even if after cluster segmentation, each attribute is all more similar between the point in class,
But since each point is at random unordered, the compression effectiveness directly mapped is very poor.Therefore, it before mapping, needs to obtain in each class
The optimal alignment sequence of all the points, to ensure to maximize the similitude of one-dimensional sequence, i.e. difference between the point of front and back is minimum.Specifically
Ground, it is assumed that a class C containing m pointj={ x1,x2,…,xm, then, when needing to find a kind of mapping of progress zigzag type
The C ' that puts in orderj={ x(1),x(2),…,x(m), so that under the sequenceValue it is minimum.
The present invention solves this problem using genetic algorithm.In the algorithm, a point in each gene representation class, one
A individual comprising m gene indicates that a kind of arrangement mode of class, multiple individuals constitute a population.It defines individual in population
Fitness isThe sum of the distance of i.e. all consecutive points is smaller, and individual fitness is higher, thus
There is more maximum probability to survive.Specific algorithm flow is as follows:
Algorithm input are as follows: the class (including m point) after cluster segmentation
Algorithm output are as follows: meet the point sequence of optimal mapping
(1) point in random alignment class generates n individual, constitutes initial population；
(2) fitness of each individual is calculated
(3) individual that all fitness in population are less than given threshold s is eliminated, and the individual of survival is added next-generation kind
Group；
(4) two individuals are randomly choosed from the individual of survival with Probability p1Hybridized, generates new individual, and be added
Next-generation population.Hybridizing method are as follows: the gene for intercepting one section of random-length from the leading portion of an individual is grafted onto second
The leading portion of body, then remove with the interception duplicate gene in part, guarantee that the gene of newly-generated individual does not repeat that (gene dosage is mutually
m)；
(5) with Probability p2Gene mutation is carried out to newly-generated individual.Gene mutation method are as follows: to two genes in individual
Carry out place-exchange；
(6) step (4) and (5) are repeated until the individual amount (n) of number and initial population individual in population of new generation
It is equal；
(7) repeat step (2) to (6) until in population the highest individual of fitness meet predetermined condition, output fitness
Highest individual；
For each class after layering segmentation, the present invention obtains such optimal mapping side to plane using genetic algorithm
Formula, then according to zigzag mode shown in Fig. 3 (a) by geological information (xyz value) of each point and colouring information (rgb value) etc.
Attribute finally carries out the regular image after mapping with image compression algorithm according to regular planar is respectively mapped to shown in Fig. 3 (b)
Compressed encoding.Shown in the overall procedure such as Fig. 1 (by taking the point cloud comprising geological information and colouring information as an example) of compression method.
Specific embodiments of the present invention are described above.It is to be appreciated that the invention is not limited to above-mentioned
Particular implementation, those skilled in the art can make a variety of changes or modify within the scope of the claims, this not shadow
Ring substantive content of the invention.In the absence of conflict, the feature in embodiments herein and embodiment can any phase
Mutually combination.
Claims (4)
1. a kind of 3D point cloud compaction coding method, which comprises the steps of:
Step S1: by the point divide into several classes in cloud, and make point either space coordinate or the color category in each class
Property is all very close；
Step S2: and then compressed encoding is carried out to each class respectively again.
2. a kind of 3D point cloud compaction coding method according to claim 1, which is characterized in that the step S1 passes through continuous
Central point is moved to the higher region of probability density and belongs to of a sort data point to find, is specifically comprised the following steps:
Step S101: input original point cloud data randomly selects a point as initial center point m in not labeled pointn；
Step S102: with mnCentered on point creation one class Ci.
Step S103: search in all input data points with mnThe distance between be less than radius r all neighborhood points, by these points
Labeled as having accessed, each point can be accessed repeatedly, and update each point by class CiThe number of access；
Step S104: m is calculatednWith the weighted mass center m of all neighborhood points searched outn+1, and central point is updated to mn+1；
Step S105: if mn+1And mnThe distance between be greater than the threshold value of setting, then return step (3)；Otherwise, iteration convergence,
Into in next step；
Step S106: if the central point after convergence is less than the threshold value of setting at a distance from the central point of existing class, by it
Merge with already present class；
Step S107: above step is repeated until all the points are all labeled；
Step S108: the number that each point is accessed by each class, class of the class for taking access times most as the point are read；
Step S109: the point cloud data after output category.
3. a kind of 3D point cloud compaction coding method according to claim 1, which is characterized in that step S2 by put cloud included
Data be mapped to rule two dimensional image, then using Image Coding Algorithms carry out compressed encoding, setting one contain m point
Class Cj={ x1, x2..., xm, find the C ' that puts in order when carrying out the mapping of zigzag typej={ x(1), x(2)..., x(m),
Made under the sequence using genetic algorithmValue it is minimum, a point in each gene representation class, one
A individual comprising m gene indicates that a kind of arrangement mode of class, multiple individuals constitute a population, defines individual in population
Fitness isThe sum of the distance of i.e. all consecutive points is smaller, and individual fitness is higher, thus
There is more maximum probability to survive.
4. according to a kind of 3D point cloud compaction coding method according to claim 3, which is characterized in that the genetic algorithm master
Want the following steps are included:
Step 201: a class after input cluster segmentation, the point in random alignment class generate n individual, composition initial population；
Step S202: the fitness of each individual is calculated
Step S203: the individual that all fitness in population are less than given threshold s is eliminated, and the next generation is added in the individual of survival
Population；
Step S204: two individuals are randomly choosed from the individual of survival with Probability p1Hybridized, generates new individual, and add
Enter next-generation population, wherein hybridizing method are as follows: the gene for intercepting one section of random-length from the leading portion of an individual is grafted onto the
Two individual leading portions, then remove and do not repeated mutually with the interception duplicate gene in part, the gene of the newly-generated individual of guarantee；
Step S205: with Probability p2Gene mutation is carried out to newly-generated individual.Gene mutation method are as follows: to two bases in individual
Because carrying out place-exchange；
Step 206: step S204 and S205 are repeated, until the individual amount of number and initial population individual in population of new generation
It is equal；
Step S207: repeat step S202 to S206 until in population the highest individual of fitness meet predetermined condition, output is suitable
The highest individual of response；
Step 208: output meets the point sequence of optimal mapping.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810322696.8A CN110363822A (en) | 2018-04-11 | 2018-04-11 | A kind of 3D point cloud compression method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810322696.8A CN110363822A (en) | 2018-04-11 | 2018-04-11 | A kind of 3D point cloud compression method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110363822A true CN110363822A (en) | 2019-10-22 |
Family
ID=68214565
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810322696.8A Pending CN110363822A (en) | 2018-04-11 | 2018-04-11 | A kind of 3D point cloud compression method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110363822A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111340899A (en) * | 2020-02-14 | 2020-06-26 | 福州大学 | Compression sampling and reconstruction method of color point cloud |
| CN111405281A (en) * | 2020-03-30 | 2020-07-10 | 北京大学深圳研究生院 | Point cloud attribute information encoding method, point cloud attribute information decoding method, storage medium and terminal equipment |
| CN111523475A (en) * | 2020-04-23 | 2020-08-11 | 江苏黑麦数据科技有限公司 | Method and device for identifying object in 3D point cloud, storage medium and processor |
| CN112995758A (en) * | 2019-12-13 | 2021-06-18 | 鹏城实验室 | Point cloud data encoding method, decoding method, storage medium, and device |
| CN113034627A (en) * | 2021-03-30 | 2021-06-25 | 太原科技大学 | Virtual structured light 3D point cloud compression method based on geometric rearrangement |
| CN113114608A (en) * | 2020-01-10 | 2021-07-13 | 上海交通大学 | Point cloud data packaging method and transmission method |
| CN113453018A (en) * | 2020-03-25 | 2021-09-28 | 浙江大学 | Point cloud attribute prediction method and device |
| WO2023045044A1 (en) * | 2021-09-27 | 2023-03-30 | 北京大学深圳研究生院 | Point cloud coding method and apparatus, electronic device, medium, and program product |
| CN117372552A (en) * | 2023-10-18 | 2024-01-09 | 南京邮电大学 | Three-dimensional point cloud data compression method of industrial product facing complex surface |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700398A (en) * | 2014-12-31 | 2015-06-10 | 西安理工大学 | Point cloud scene object extracting method |
| CN104750854A (en) * | 2015-04-16 | 2015-07-01 | 武汉海达数云技术有限公司 | Mass three-dimensional laser point cloud compression storage and rapid loading and displaying method |
| CN105630905A (en) * | 2015-12-14 | 2016-06-01 | 西安科技大学 | Scattered-point cloud data based hierarchical compression method and apparatus |
| CN106845399A (en) * | 2017-01-18 | 2017-06-13 | 北京林业大学 | A kind of method that use hierarchical cluster mode extracts individual tree information from LiDAR point cloud |
| WO2017126314A1 (en) * | 2016-01-22 | 2017-07-27 | Mitsubishi Electric Corporation | Method for compressing point cloud |
-
2018
- 2018-04-11 CN CN201810322696.8A patent/CN110363822A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104700398A (en) * | 2014-12-31 | 2015-06-10 | 西安理工大学 | Point cloud scene object extracting method |
| CN104750854A (en) * | 2015-04-16 | 2015-07-01 | 武汉海达数云技术有限公司 | Mass three-dimensional laser point cloud compression storage and rapid loading and displaying method |
| CN105630905A (en) * | 2015-12-14 | 2016-06-01 | 西安科技大学 | Scattered-point cloud data based hierarchical compression method and apparatus |
| WO2017126314A1 (en) * | 2016-01-22 | 2017-07-27 | Mitsubishi Electric Corporation | Method for compressing point cloud |
| CN106845399A (en) * | 2017-01-18 | 2017-06-13 | 北京林业大学 | A kind of method that use hierarchical cluster mode extracts individual tree information from LiDAR point cloud |
Non-Patent Citations (4)
| Title |
|---|
| SHALINI SINGH 等: "Study of Variation in TSP using Genetic Algorithm and Its Operator Comparison", 《INTERNATIONAL JOURNAL OF SOFT COMPUTING AND ENGINEERING (IJSCE)》 * |
| WENJIE ZHU 等: "Lossless Point Cloud Geometry Compression via Binary Tree Partition and Intra Prediction", 《2017 IEEE 19TH INTERNATIONAL WORKSHOP ON MULTIMEDIA SIGNAL PROCESSING (MMSP)》 * |
| ZHANG XIMIN 等: "Six dimensional clustering segmentation of color point cloud", 《2016 INTERNATIONAL CONFERENCE ON AUDIO, LANGUAGE AND IMAGE PROCESSING (ICALIP)》 * |
| 禹永萍: "基于深度图像的三维重建技术研究", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (18)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112995758A (en) * | 2019-12-13 | 2021-06-18 | 鹏城实验室 | Point cloud data encoding method, decoding method, storage medium, and device |
| CN112995758B (en) * | 2019-12-13 | 2024-02-06 | 鹏城实验室 | Encoding method, decoding method, storage medium and equipment for point cloud data |
| JP7577357B2 (en) | 2020-01-10 | 2024-11-05 | 上海交通大学 | Method for encapsulating and transmitting point cloud data |
| US12056091B2 (en) | 2020-01-10 | 2024-08-06 | Shanghai Jiao Tong University | Point cloud data encapsulation method and point cloud data transmission method |
| CN113114608A (en) * | 2020-01-10 | 2021-07-13 | 上海交通大学 | Point cloud data packaging method and transmission method |
| CN113114608B (en) * | 2020-01-10 | 2022-06-10 | 上海交通大学 | Point cloud data packaging method and transmission method |
| JP2023510822A (en) * | 2020-01-10 | 2023-03-15 | 上海交通大学 | Point cloud data encapsulation method and transmission method |
| CN111340899A (en) * | 2020-02-14 | 2020-06-26 | 福州大学 | Compression sampling and reconstruction method of color point cloud |
| CN111340899B (en) * | 2020-02-14 | 2022-09-06 | 福州大学 | Compression sampling and reconstruction method of color point cloud |
| CN113453018A (en) * | 2020-03-25 | 2021-09-28 | 浙江大学 | Point cloud attribute prediction method and device |
| CN113453018B (en) * | 2020-03-25 | 2022-05-10 | 浙江大学 | Point cloud attribute value prediction method and device |
| CN111405281A (en) * | 2020-03-30 | 2020-07-10 | 北京大学深圳研究生院 | Point cloud attribute information encoding method, point cloud attribute information decoding method, storage medium and terminal equipment |
| CN111523475B (en) * | 2020-04-23 | 2023-12-19 | 江苏黑麦数据科技有限公司 | Method and device for identifying object in 3D point cloud, storage medium and processor |
| CN111523475A (en) * | 2020-04-23 | 2020-08-11 | 江苏黑麦数据科技有限公司 | Method and device for identifying object in 3D point cloud, storage medium and processor |
| CN113034627A (en) * | 2021-03-30 | 2021-06-25 | 太原科技大学 | Virtual structured light 3D point cloud compression method based on geometric rearrangement |
| WO2023045044A1 (en) * | 2021-09-27 | 2023-03-30 | 北京大学深圳研究生院 | Point cloud coding method and apparatus, electronic device, medium, and program product |
| CN117372552A (en) * | 2023-10-18 | 2024-01-09 | 南京邮电大学 | Three-dimensional point cloud data compression method of industrial product facing complex surface |
| CN117372552B (en) * | 2023-10-18 | 2024-07-05 | 南京邮电大学 | Three-dimensional point cloud data compression method of industrial product facing complex surface |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110363822A (en) | A kind of 3D point cloud compression method | |
| CN108920720B (en) | Large-scale image retrieval method based on depth hash and GPU acceleration | |
| Liu et al. | Scene classification using hierarchical Wasserstein CNN | |
| Liu et al. | Fast detection of dense subgraphs with iterative shrinking and expansion | |
| US6075538A (en) | Time and space efficient data structure and method and apparatus for using the same for surface rendering | |
| RU2674326C2 (en) | Method of formation of neural network architecture for classification of object taken in cloud of points, method of its application for teaching neural network and searching semantically alike clouds of points | |
| CN111507415B (en) | Multi-source atmosphere data clustering method based on distribution density | |
| CN109979008A (en) | A kind of point cloud band division methods based on attribute | |
| CN104036012A (en) | Dictionary learning method, visual word bag characteristic extracting method and retrieval system | |
| CN114820975B (en) | Three-dimensional scene simulation reconstruction system and method based on all-element parameter symbolization | |
| CN113487523B (en) | Method and device for optimizing graph contour, computer equipment and storage medium | |
| CN109271441A (en) | A kind of visualization clustering method of high dimensional data and system | |
| CN113094463A (en) | Unstructured point cloud storage method, device, equipment and medium | |
| CN113628329B (en) | Zero-sample sketch three-dimensional point cloud retrieval method | |
| CN111680579A (en) | Remote sensing image classification method for adaptive weight multi-view metric learning | |
| CN111860359B (en) | Point cloud classification method based on improved random forest algorithm | |
| CN110097581B (en) | Method for constructing K-D tree based on point cloud registration ICP algorithm | |
| CN105138607B (en) | A kind of KNN querying methods based on combination grain distributed memory grid index | |
| CN104778212B (en) | Map datum generation method and device, map datum read method and device | |
| CN115408382A (en) | Vector space information gridding storage management method for large-scale satellite data | |
| Elkin | A new compressed cover tree for k-nearest neighbour search and the stable-under-noise mergegram of a point cloud | |
| CN114359902A (en) | Three-dimensional point cloud semantic segmentation method based on multi-scale feature fusion | |
| CN113256793A (en) | Three-dimensional data processing method and system | |
| CN115952252B (en) | Semantic tile data processing method and device based on dynamic rendering and electronic equipment | |
| CN117456190A (en) | Point cloud semantic segmentation method for large-scale complex scene |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191022 |