CN109636844A - A method of the complicated desktop point cloud segmentation based on 3D bilateral symmetry - Google Patents
A method of the complicated desktop point cloud segmentation based on 3D bilateral symmetry Download PDFInfo
- Publication number
- CN109636844A CN109636844A CN201811372772.2A CN201811372772A CN109636844A CN 109636844 A CN109636844 A CN 109636844A CN 201811372772 A CN201811372772 A CN 201811372772A CN 109636844 A CN109636844 A CN 109636844A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- symmetrical
- cloud
- point
- bilateral symmetry
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/68—Analysis of geometric attributes of symmetry
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T15/00—3D [Three Dimensional] image rendering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Abstract
A method of the complicated desktop point cloud segmentation based on 3D bilateral symmetry belongs to computer picture visual field.Pretreatment is simplified to the desktop complexity point cloud of acquisition first, then by extracting the boundary curve with the surface normal in match point cloud, the 3D bilateral symmetry in scene is effectively detected, it establishes symmetrical hypothesis and initializes cutting procedure using symmetrical hypothesis, finally, carrying out point cloud segmentation according to bilateral symmetry constrained attributes.The experimental results showed that method of the invention can be complete by desktop complexity point cloud segmentation, and make point cloud segmentation more accurate.
Description
Technical field
The invention belongs to computer picture visual field, be related to it is a kind of using geometrical constraint to complicated desktop point cloud minute
The method cut.
Background technique
Object segmentation identification under three-dimensional environment, is one of indoor moving service robot important research basic content, right
The grasping body of view-based access control model and manipulation are extremely important；Therefore, the segmentation under three-dimensional environment is identified to pass
It is important, be a challenging task, or even by some researchers be considered one can not define the problem of.
Traditional dividing method be based on more RGB or gray level image analysis, and combine texture, color and image outline feature into
Row grouping.The method of currently used three-dimensional segmentation mainly has the method based on boundary, the method based on region, is belonged to based on cluster
The method of property, based on model and the method based on figure etc..These algorithms are suitable for certain segmentation tasks, but in complex object
Robust division object is difficult in living environment.
The dividing method of traditional desktop complexity point cloud mainly has the following insufficient:
(1) very sensitive to noise spot based on the region segmentation method on side, robustness is poor, is easy the extension side according to mistake
It to being tracked, and cannot be guaranteed that obtained boundary line forms closed ring, to be unable to complete region segmentation, cause to divide
Not exclusively；
(2) dividing method based on face is the problem is that be difficult to select suitable seed point and smooth boundary, method
Validity dependent on complicated judgement control；
(3) method based on cluster attribute will directly determine the classification number and curved surface of curved surface for complicated curved surface
Type is very difficult, carries out secondary treatment to the dough sheet fine crushing being easy to appear and also increases the difficulty of algorithm.
Summary of the invention
In order to overcome the above-mentioned shortcomings of the prior art, the present invention proposes a kind of bis- based on 3D on the basis of 3D bilateral symmetry
The method of the symmetrical complicated desktop point cloud segmentation in side.
The present invention is achieved by the following technical solutions.
A kind of method of complicated desktop point cloud segmentation based on 3D bilateral symmetry of the present invention, comprising the following steps:
(1) point cloud boundary is extracted, detecting symmetrical boundary curve, to carry out match point cloud symmetrical；
(2) main feature for cloud distribution is found, and uses its maximum value as the symmetrical plane of two curves of alignment, it should
Process will be run on each pair of boundary curve, will generate a series of 3D bilateral symmetry hypothesis of plane mechanism set S to input point cloud S；
(3) according to the symmetrical hypothesis set of (2), using 3D bilateral symmetry constrained attributes to Desktop-scene complexity point Yun Jinhang
Whether segmentation, test point meet symmetrical hypothesis set S；
(4) the invalid symmetrical hypothesis collection of removal, using ICP registration Algorithm, obtained cut-point cloud H and original point cloud S are matched
Standard, in order to ensure restraining fast and accurately, point of use respective distances measure to carry out correspondence estimation, and force matching be
Point cloud after being divided correspondingly.
Symmetrical hypothesis set S described in step (2) of the present invention, in order to which whether test point cloud midpoint meets symmetrical hypothesis collection
It closes, is divided into three kinds of situations: 1) putting cloud p just and put symmetrical corresponding on cloud P；2) point cloud p is symmetrically corresponded to the position being blocked,
It does not support symmetrically but meets in cloud P；3) point cloud p symmetrically corresponds to a non-confined space, then is not belonging to assume set
S finds corresponding symmetrical hypothesis collection, i.e. point in symmetrical border curve by three kinds of above-mentioned symmetric cases.
After finding symmetrical hypothesis set S, when object segmentation, first have to retain in the point cloud of symmetrical consistency, no
The point cloud removal for meeting symmetrical consistency, for all the points cloud object in split sence, needs to establish the prospect based on multiple figures
Segmentation problem, for each symmetrical one figure of hypothesis construction, interior joint corresponds between the consecutive points of point cloud and symmetrically adjacent
Side is established between point, if ξ={ fg,bgThe label of object and background is corresponded to respectively.By finding label fp∈ ξ distribution
Carry out cutting object to the label f of all the points in cloud to assume.
The method usefulness of the invention is that the invention belongs to the dividing method of the geometric attribute using object, is not only fitted
It is also used for rigid body and non-rigid is as long as object has 3D bilateral symmetry attribute can obtain good segmentation effect.3D is bilateral
The combination of symmetric properties and point cloud segmentation, so that complicated object on table top segmentation becomes to change substantially.Antificielle object all has substantially
Have 3D bilateral symmetry attribute, according to this, 3D bilateral symmetry constraint combination make the method for the invention do not need any point,
The geometric attribute constraint of the priori knowledge in line or face, 3D bilateral symmetry can distinguish desktop complexity point cloud, can be complete
At segmentation.The combination of ICP registration Algorithm, so that point cloud segmentation is more accurate.
Detailed description of the invention
Fig. 1 is flow chart of the present invention.
Fig. 2 is 3D bilateral symmetry detection schematic diagram.
Fig. 3 is a cloud plane monitoring-network schematic diagram.
Fig. 4 is experimental result of the invention under the cloud file of difference.Wherein, left column is input point cloud, and the right side is classified as segmentation knot
Fruit.
Specific embodiment
A specific embodiment of the invention is described with reference to the accompanying drawing, preferably so as to those skilled in the art
Understand the present invention.
Fig. 1 is flow chart of the present invention.
In the present embodiment, as shown in Figure 1, the present invention include the steps that four it is main: point cloud pretreatment, extract point a cloud side
Boundary, detection object point cloud be symmetrical, complicated point cloud segmentation.Entire method is the desktop complexity point cloud for inputting Kinect sensor acquisition
Image exports as the point cloud file after segmentation.
In step 1, the desktop complexity point cloud chart picture of input Kinect sensor acquisition, using voxel filter to a cloud
Data carry out down-sampling, simplify cloud quantity while guaranteeing point cloud feature.Then test point cloud boundary point and line are side
Boundary's curve.Symmetrically assume to collect according to the symmetrical border curve construction of detection, by decision-point cloud midpoint whether in symmetrical hypothesis collection
Middle cut-point cloud is completed last point cloud segmentation finally by removal null hypothesis collection and ICP point cloud registering and is exported.
Each step is described in detail below.
1,3D symmetric points detect.
The bilateral symmetry detection of three-dimensional data needs to lead to by needing the orientation point in three-dimensional data with symmetrical consistency
Cross normal n a littlesWith point and point distance dsTo describe bilateral symmetry plane S={ ns,ds}.Point p and its surface normal vector n
Associated symmetrical corresponding points pr, symmetrical normal nr, solution is expressed as follows:
pr=p-2ns(p·ns-ds)
nr=n-2ns(n·ns) (1)
Symmetrical matching degree between two points can by the normal of a point and another about the symmetrical of symmetrical plane S
The differential seat angle of normal is measured:
It is as shown in Figure 2 to solve schematic diagram.
2, point cloud surface normal Boundary Extraction with match.
Using the interactively pick-up target signature Boundary algorithm on the dispersion point cloud with multiple features, is established and dissipated using KD tree
The disorderly spatial topotaxy of point cloud, calculates the k neighborhood of each data point.The least square being made up of data point k neighborhood is flat
The subpoint differential seat angle in face judges whether the point is boundary point according to the maximum value of differential seat angle.Then minimum spanning tree is used
Boundary point is joined together to form one group of disjoint boundary curve C={ C by algorithm1、C2……Cn}.It is extracted in above-mentioned algorithm
It is symmetrical come match point cloud by finding symmetrical boundary curve after boundary curve.The symmetrical algorithm of match point cloud is: given a pair
Boundary curve, by the way that threshold value M is arrangeds(pc1,pc2) filter invalid matching.
In order to restore global symmetrical plane, using the similar approach proposed with document [14], plane to be restored is considered as entirely
The problem of office's symmetrical plane probability density function sample, recovery overall situation symmetrical plane, is equivalent to carry out these planes to be restored
Cluster.A main feature for cloud distribution is found by Mean Shift cluster, and uses its maximum value as two curves of alignment
Symmetrical plane.The process will be run on each pair of boundary curve, and it is flat to generate a series of 3D bilateral symmetries to input point cloud S
Assume set in face.
3, cloud symmetry division is put.
The partitioning algorithm key is to find corresponding symmetrical hypothesis collection, i.e. point in symmetrical border curve, energy letter
Number as shown in Equation 3, wherein D (p) be symmetrical consistent detection,For smooth item,For symmetrical edge limit entry.
(1) enabling D (p) is symmetrical consistent item, it assures that being marked as prospect with consistent point is currently symmetrically assumed.Given one
A symmetrical hypothesis S, the symmetrical corresponding points p of a point pr.If it is less than threshold value | | pr- p'| | < dmax, then point p belongs to object point
Point in cloud, therefore its background weight is set as 0.Prospect weight is according between symmetrical corresponding normal and symmetrical adjacent normal
Angle setting:
On the other hand, if the distance to nearest field is greater than dmax, then point p does not have symmetrical field, and its prospect
Weight is arranged to 0.If symmetrical corresponding points occupy the space being blocked, background weight is arranged to 0.If it reflects
To the space not being blocked, then its background weight by with to being proportionally arranged at a distance from closest approach:
(2) it enablesIt is expressed as smooth item.Its meeting forced line segment boundary is placed along curved surface normal side, without more than limit
Determine the curved surface of threshold value.Consecutive points N in point cloudsmoothIt is to be estimated by the way that each point is connected to its immediate 5 consecutive points
's.Weight criterion is adjusted based on convexity standard, gives two o'clock p1And p2If n1·(p1-p2) > 0, they are defined as by we
Convex arrangement.Binary weights setting between them are as follows:
(3) it enablesIt is expressed as symmetrical edge limit entry, establishes symmetrical neighboring edge between cloud all the points, it may be assumed that Nsym
=p, p'| | pr- p'| | < dmax, weight:
For the ordinary skill in the art, introduction according to the present invention, do not depart from the principle of the present invention with
In the case where spirit, changes, modifications that embodiment is carried out, replacement and variant still fall within protection scope of the present invention it
It is interior.
Experiment: the experiment porch that this experiment uses is Inter Core i5-6400@2.3GHz, 8GB memory Ubuntu
16.04 operating systems, open source point cloud library PCL1.8.0, using desktop point cloud data library in Maryland university room, experimental result
As shown in Figure 4.
Shown in experimental result, in the first row, the transition between loudspeaker and coffee cartridge is non-recessed.And inventive algorithm
By symmetry constraint, can identify well between transition be not concavity and convexity object.In a second row, Kinect box is by milk
Box stops, and inventive algorithm then identifies its different object well, and the box being blocked also can be identified correctly.Below
When the point cloud object of 2 rows complexity, symmetry provides the group forming criterion of global object's rank, this enables method of the invention
These scenes of correct Ground Split.
Conclusion: experimental data of the invention is to grind with the three dimensional point cloud of the PCD format scanned of Microsoft Kinect 2.0
Study carefully object, desktop complexity point cloud is split using the constraint of 3D bilateral symmetry.The method of the present invention does not need to have indoor object
Priori understanding, only depends on the geological information of object, constrains segmentation object by 3D bilateral symmetry.The experimental results showed that the party
Method has certain superiority and advance in complexity point cloud identification.
Claims (1)
1. a kind of method of the complicated desktop point cloud segmentation based on 3D bilateral symmetry, it is characterized in that the following steps are included:
(1) point cloud boundary is extracted, detecting symmetrical boundary curve, to carry out match point cloud symmetrical；
(2) main feature for cloud distribution is found, and uses its maximum value as the symmetrical plane of two curves of alignment, the process
It will be run on each pair of boundary curve, a series of 3D bilateral symmetry hypothesis of plane mechanism set S will be generated to input point cloud S；
(3) according to the symmetrical hypothesis set of (2), using 3D bilateral symmetry constrained attributes to Desktop-scene complexity point cloud minute
It cuts, whether test point meets symmetrical hypothesis set S；
(4) the invalid symmetrical hypothesis collection of removal, using ICP registration Algorithm, obtained cut-point cloud H is registrated with original point cloud S, is
Ensure to restrain fast and accurately, point of use respective distances are measured carrying out correspondence estimation, and forcing matching is one by one
It is corresponding divided after point cloud；
Symmetrical hypothesis set S described in step (2) is divided into three kinds in order to which whether test point cloud midpoint meets symmetrical hypothesis set
Situation: 1) it puts cloud p just and puts symmetrical corresponding on cloud P；2) point cloud p is symmetrically corresponded to the position being blocked, and is not supported symmetrical
But meet in cloud P；3) point cloud p symmetrically corresponds to a non-confined space, then is not belonging to assume set S, by above-mentioned
Three kinds of symmetric cases, find and corresponding symmetrical assume collection, i.e. point in symmetrical border curve；
Find it is symmetrical assume set S after, when object segmentation, first by the point cloud reservation of symmetrical consistency, be unsatisfactory for pair
The point cloud removal for claiming consistency, for all the points cloud object in split sence, needs to establish the foreground segmentation based on multiple figures and asks
Topic symmetrically assumes that one figure of construction, interior joint correspond between the consecutive points of a cloud and between symmetrical consecutive points to be each
Side is established, if ξ={ fg,bgThe label of object and background is corresponded to respectively；By finding label fp∈ ξ distributes to a cloud
In all the points label f come cutting object hypothesis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811372772.2A CN109636844B (en) | 2018-11-19 | 2018-11-19 | Complex desktop point cloud segmentation method based on 3D bilateral symmetry |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811372772.2A CN109636844B (en) | 2018-11-19 | 2018-11-19 | Complex desktop point cloud segmentation method based on 3D bilateral symmetry |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109636844A true CN109636844A (en) | 2019-04-16 |
| CN109636844B CN109636844B (en) | 2023-05-02 |
Family
ID=66068413
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811372772.2A Active CN109636844B (en) | 2018-11-19 | 2018-11-19 | Complex desktop point cloud segmentation method based on 3D bilateral symmetry |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109636844B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116878396A (en) * | 2023-09-06 | 2023-10-13 | 国网山西省电力公司超高压输电分公司 | Sag measurement method and system based on remote laser |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060078172A1 (en) * | 2004-06-03 | 2006-04-13 | Arizona Board Of Regents, A Body Corporate Of The State Of Arizona | 3D face authentication and recognition based on bilateral symmetry analysis |
| CN102188311A (en) * | 2010-12-09 | 2011-09-21 | 南昌大学 | Embedded visual navigation control system and method of intelligent wheelchair |
| KR20130037996A (en) * | 2011-10-07 | 2013-04-17 | 한국전자통신연구원 | Auto segmentation method of 3d point clouds |
| CN105190689A (en) * | 2013-06-14 | 2015-12-23 | 英特尔公司 | Image processing including adjoin feature based object detection, and/or bilateral symmetric object segmentation |
| CN106408581A (en) * | 2016-09-12 | 2017-02-15 | 厦门大学 | Rapid three-dimensional point cloud linear extraction method |
| CN106780751A (en) * | 2017-01-19 | 2017-05-31 | 桂林电子科技大学 | Three-dimensional point cloud method for reconstructing based on improved shielding Poisson algorithm |
| CN106846484A (en) * | 2017-02-20 | 2017-06-13 | 深圳市唯特视科技有限公司 | A kind of food volume method of estimation based on dual-view three-dimensional reconstruction |
| US20190271858A1 (en) * | 2017-01-27 | 2019-09-05 | Carl Zeiss Vision International Gmbh | Method, apparatus, and computer program for establishing a representation of a spectacle lens edge |
-
2018
- 2018-11-19 CN CN201811372772.2A patent/CN109636844B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20060078172A1 (en) * | 2004-06-03 | 2006-04-13 | Arizona Board Of Regents, A Body Corporate Of The State Of Arizona | 3D face authentication and recognition based on bilateral symmetry analysis |
| CN102188311A (en) * | 2010-12-09 | 2011-09-21 | 南昌大学 | Embedded visual navigation control system and method of intelligent wheelchair |
| KR20130037996A (en) * | 2011-10-07 | 2013-04-17 | 한국전자통신연구원 | Auto segmentation method of 3d point clouds |
| CN105190689A (en) * | 2013-06-14 | 2015-12-23 | 英特尔公司 | Image processing including adjoin feature based object detection, and/or bilateral symmetric object segmentation |
| CN106408581A (en) * | 2016-09-12 | 2017-02-15 | 厦门大学 | Rapid three-dimensional point cloud linear extraction method |
| CN106780751A (en) * | 2017-01-19 | 2017-05-31 | 桂林电子科技大学 | Three-dimensional point cloud method for reconstructing based on improved shielding Poisson algorithm |
| US20190271858A1 (en) * | 2017-01-27 | 2019-09-05 | Carl Zeiss Vision International Gmbh | Method, apparatus, and computer program for establishing a representation of a spectacle lens edge |
| CN106846484A (en) * | 2017-02-20 | 2017-06-13 | 深圳市唯特视科技有限公司 | A kind of food volume method of estimation based on dual-view three-dimensional reconstruction |
Non-Patent Citations (3)
| Title |
|---|
| 刘继忠 等: "基于双边对称约束的复杂点云分割" * |
| 朱晓鑫;谢翔;李国林;: "一种大场景有序点云的快速、准确分割方法" * |
| 郭浩: "动物体表三维数据获取与处理算法研究" * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116878396A (en) * | 2023-09-06 | 2023-10-13 | 国网山西省电力公司超高压输电分公司 | Sag measurement method and system based on remote laser |
| CN116878396B (en) * | 2023-09-06 | 2023-12-01 | 国网山西省电力公司超高压输电分公司 | Sag measurement method and system based on remote laser |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109636844B (en) | 2023-05-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| He et al. | Food image analysis: Segmentation, identification and weight estimation | |
| Kong et al. | Dietcam: Regular shape food recognition with a camera phone | |
| CN109325998A (en) | A kind of indoor 3D modeling method, system and relevant apparatus based on point cloud data | |
| CN106845372A (en) | The ship target detection recognition method and system of a kind of space remote sensing optical imagery | |
| CN107239777B (en) | Tableware detection and identification method based on multi-view graph model | |
| CN103337072A (en) | Texture and geometric attribute combined model based indoor target analytic method | |
| CN109035207B (en) | Density self-adaptive laser point cloud characteristic detection method | |
| CN103366387A (en) | Selecting between clustering techniques for displaying images | |
| Pahwa et al. | Locating 3D object proposals: A depth-based online approach | |
| Keskin et al. | Image classification of human carcinoma cells using complex wavelet-based covariance descriptors | |
| Ecins et al. | Cluttered scene segmentation using the symmetry constraint | |
| Stucker et al. | Supervised outlier detection in large-scale MVS point clouds for 3D city modeling applications | |
| Ecins et al. | Seeing behind the scene: Using symmetry to reason about objects in cluttered environments | |
| Chowdhary et al. | Three-dimensional object recognition based intelligence system for identification | |
| CN109636844A (en) | A method of the complicated desktop point cloud segmentation based on 3D bilateral symmetry | |
| Yao | Image segmentation based on Sobel edge detection | |
| Cui et al. | Global propagation of affine invariant features for robust matching | |
| CN109993151A (en) | A kind of 3 D video visual attention detection method based on the full convolutional network of multimode | |
| Hristov et al. | A software system for classification of archaeological artefacts represented by 2D plans | |
| Dimiccoli et al. | Exploiting t-junctions for depth segregation in single images | |
| CN104268571B (en) | A kind of Infrared Multi-Target dividing method based on minimum tree cluster | |
| Boukamcha et al. | 3D face landmark auto detection | |
| Grum et al. | 3D modeling of multiple-object scenes from sets of images | |
| CN106340038B (en) | A kind of depth order inference method based on object in monocular image | |
| CN111462123B (en) | Point cloud data segmentation method based on spectral clustering |
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 |