CN110009671A - A kind of grid surface reconstructing system of scene understanding - Google Patents
A kind of grid surface reconstructing system of scene understanding Download PDFInfo
- Publication number
- CN110009671A CN110009671A CN201910131449.4A CN201910131449A CN110009671A CN 110009671 A CN110009671 A CN 110009671A CN 201910131449 A CN201910131449 A CN 201910131449A CN 110009671 A CN110009671 A CN 110009671A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- unit
- data
- segmentation
- cloud
- 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
- 230000011218 segmentation Effects 0.000 claims abstract description 91
- 238000013507 mapping Methods 0.000 claims abstract description 43
- 238000012545 processing Methods 0.000 claims abstract description 25
- 239000013598 vector Substances 0.000 claims description 43
- 230000000007 visual effect Effects 0.000 claims description 38
- 238000000034 method Methods 0.000 claims description 29
- 230000008439 repair process Effects 0.000 claims description 21
- 230000002159 abnormal effect Effects 0.000 claims description 14
- 230000004438 eyesight Effects 0.000 claims description 14
- 238000011017 operating method Methods 0.000 claims description 2
- 238000007781 pre-processing Methods 0.000 claims description 2
- 238000004422 calculation algorithm Methods 0.000 abstract description 23
- 238000004458 analytical method Methods 0.000 abstract description 19
- 230000016776 visual perception Effects 0.000 abstract description 17
- 238000013145 classification model Methods 0.000 abstract description 2
- 238000007405 data analysis Methods 0.000 abstract description 2
- 230000006870 function Effects 0.000 description 32
- 238000013461 design Methods 0.000 description 11
- 239000000284 extract Substances 0.000 description 11
- 238000000605 extraction Methods 0.000 description 10
- 238000005259 measurement Methods 0.000 description 9
- 244000062793 Sorghum vulgare Species 0.000 description 8
- 235000019713 millet Nutrition 0.000 description 8
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 7
- 230000008447 perception Effects 0.000 description 7
- 230000009467 reduction Effects 0.000 description 6
- 230000003287 optical effect Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 238000003384 imaging method Methods 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 3
- 238000000513 principal component analysis Methods 0.000 description 3
- 238000005070 sampling Methods 0.000 description 3
- 239000007787 solid Substances 0.000 description 3
- 238000000638 solvent extraction Methods 0.000 description 3
- 241000350052 Daniellia ogea Species 0.000 description 2
- 241000404095 Heteranthemis Species 0.000 description 2
- 208000016063 arterial thoracic outlet syndrome Diseases 0.000 description 2
- 238000009795 derivation Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 238000005303 weighing Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T3/00—Geometric image transformations in the plane of the image
- G06T3/04—Context-preserving transformations, e.g. by using an importance map
-
- 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
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
-
- 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
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of grid surface reconstructing systems of scene understanding, belong to three-dimensional graphics field, and present system includes data input cell, data processing unit, curve reestablishing unit, point cloud network surface model output unit;The data processing unit includes point cloud pretreatment unit, point cloud segmentation unit, point cloud reparation unit;The curve reestablishing unit includes hypersphere map unit, binary segmentation reconstruction unit;Complex scene divide based on the junior unit with local visual perception characteristics by unit of the invention first, analyse in depth space structure, the geometric shape information of sub- cut zone, construct corresponding self study classification model, view-based access control model hypersphere mapping principle, put forward a whole set of data analysis and high-quality surface Model Reconstruction algorithm with underlying semantics characteristic, ρ-ε when meeting using geometrical reconstruction rebuilds criterion.
Description
Technical field
The invention belongs to three-dimensional graphics fields, the present invention relates to a kind of grid surface method for reconstructing of scene understanding and are
System, the grid surface method for reconstructing of specifically a kind of scene understanding.
Background technique
For measurement institute's invocation point cloud big data quantity, unordered, localized loss at random the characteristics of, how to multiple types
The tested entity of (such as: LIDAR data, the data that the scanners such as ATOS structure light, Faro laser obtain) carries out efficiently, accurately
Its corresponding triangle grid model is reconstructed, is always the hot spot of educational circles's research.
Comprehensive researching and analysing in terms of algorithm for reconstructing theory, model origin classification, mode three of processing massive point cloud
From the point of view of, algorithm for reconstructing when with the characteristic of following three aspects, will enable algorithm for reconstructing complicated form, a variety of is effectively treated
The mass cloud data in source: (a) with the segmentation towards broad sense geometric graphic element of certain visual perception;(b) there is certain field
Automatically it repairs in the localized loss region that scape understands;(c) following ρ-ε criterion, there is whole and incremental nature quasi- computational geometry to rebuild mould
Type.
For measurement institute's invocation point cloud big data quantity, unordered, localized loss at random the characteristics of, how to multiple types
The tested entity of (such as: LIDAR data, the data that the scanners such as ATOS structure light, Faro laser obtain) carries out efficiently, accurately
Its corresponding triangle grid model is reconstructed, is always the hot spot of educational circles's research.The documents and materials, soft published from the country
From the point of view of part system research achievement, for the point cloud model for being originated from different scenes, still lack to being provided simultaneously with " Grid-oriented curved surface weight
The point cloud segmentation built ", " absent region understood with model scene is automatically repaired ", " the point Yun Qu with computational geometry characteristic
Systematicness in terms of the point cloud grid surface algorithm for reconstructing of three characteristics of face reconstruction theory " is realized.
In addition, with the raising of measuring device precision, efficiency, vertex that the point cloud model of acquisition is included is measured mostly with thousand
Ten thousand, it hundred million counts.Since point cloud model is sampled to the Direct Three-dimensional of mock-up or scene, the Limited information contained, not
The configuration of surface that can accurately indicate tested entity, to carry out more accurate digital expression to mock-up, and carries out height
The subsequent operations such as modeling, editor, analysis, the rendering of effect need to carry out the Surface Reconstruction based on triangle gridding to cloud.Cause
This, research of the invention is come into being.
Summary of the invention
Aiming at the problems existing in the prior art, the present invention provides a kind of for come the point cloud that is originated from different scenes the present invention
Data, simulation and abstract visual imaging theory, it then follows ρ-ε rebuild criterion, independent of sampling density mesh reconstruction method and
System.
The present invention is implemented as follows:
A kind of grid surface reconstructing system of scene understanding, which is characterized in that the system include data input cell,
Data processing unit, curve reestablishing unit, point cloud network surface model output unit;The data processing unit includes point cloud
Pretreatment unit, point cloud segmentation unit, point cloud repair unit;The curve reestablishing unit includes hypersphere map unit, two
Member segmentation reconstruction unit.
For pre-processing to the point cloud data of input unit, removal noise is calculated and is adjusted data processing unit therein
Whole normal vector repairs absent region.Point cloud denoising is the point cloud 3-D data set to input unit, using subspace clustering
Method is known, according to the analysis of optimization function as a result, automatic other abnormal data, removal noise spot cloud;The point cloud pretreatment
It includes the calculating of cloud law vector and a consistency adjustment in unit, is used to the point cloud after disorder data recognition, noise remove
The cluster subspace of local optimum carries out curved surface extraction and hair vector calculates, consistency adjustment, be low latitudes characteristic information analysis,
Visual segments provide accurate method and swear information.It include the segmentation of point cloud " block " body and point cloud " piece " in the point cloud segmentation unit
Shape region segmentation;Point cloud " block " body segmentation, is the side scanned using frame center's curve to the regional area for having obvious skeleton
Formula, extraction obtain visual perception apparent " minimum " and change stub area;Described cloud " piece " body segmentation is to progress block segmentation
Left point cloud 3-D data set afterwards carries out the segmentation mapped based on depth map from multiple and different visual field directions, in conjunction with " block "
Segmentation result, finally obtain the segmentation result with visual perception to entire point cloud model;Point cloud, which is repaired in unit, includes
Scene understanding and classification are automatically repaired;Wherein " scene understanding " is to after segmentation, having basic " block ", " piece " shape
Simple shape point cloud subset carry out self study and return according to the relationship of closing in its topology, geometrical feature, physical space
Class, combination;Being automatically repaired for point cloud segmentation unit refers to " classification of missing data is automatically repaired ", after progress " scene understanding "
Segmentation power supply, according to scene understanding as a result, extract segmentation result in close form solid, it is close for form
Object be automatically repaired based on the image repair algorithm of template matching;Missing with bright stub area can be used
The mode that section is scanned realizes reparation automatically;It can be by the way of element figure logical operation for the structurings such as CAD point cloud
It is automatically repaired.
The curve reestablishing unit, on the basis of carrying out a data processing to cloud, segmentation, self study sorted out, to point
It cuts the simple broad sense geometric graphic element of each of result and uses the reconstruction mapped based on binary visual angle, and melt between pel
It closes, that is, is able to achieve the fine granularity geometrical reconstruction to whole object.
Further, the data that the data input cell is received are the 3-D data sets of a cloud, format include asc,
Vtx, pcd, pts, source include optics, non-optical scanner.
Further, the data input cell is used to input the 3-D data set to curve reestablishing point cloud;The number
It is used to carry out point cloud pretreatment, point cloud segmentation, point Yun Xiufu to input point cloud according to processing unit;The point cloud pretreatment unit
For to a cloud denoising, law vector calculating and consistency adjustment;The point cloud segmentation unit is used to carry out block, piece to cloud
The segmentation of body;The point cloud repairs unit and is used to repair cloud progress scene self study understanding, the classification of missing data automatically
It is multiple;The curve reestablishing unit be used for cloud carry out based on hypersphere map vision rely on whole body reconstruction, neighborhood according to
Bad increment is rebuild;The hypersphere map unit is used to convert hypersphere for the reconstruction of point cloud local visual angle visibility region
The convex closure of mapping point cloud solves;The binary segmentation reconstruction unit for realizing to broad sense geometric graphic element based on positive and negative two
The convex closure at first visual angle is rebuild;The point cloud grid model output unit is used to export the grid model after rebuilding.
Further, the operating procedure of the data processing is point cloud pretreatment, point cloud segmentation, point Yun Xiufu, right respectively
It should be the point cloud pretreatment unit, point cloud segmentation unit, point cloud and repair unit;The point cloud pretreatment unit, point cloud
It is optional to repair unit;It is specific as follows:
If the point cloud data of input has been treated via other software or tool, the point cloud pretreatment
The operation that unit, point cloud repair unit is omitted;
If the point cloud data of input is simple basic geometric graphic element, the point cloud segmentation unit can be omitted.
Further, the point cloud data of the input be treated via other software or tool, wherein
Processing include: abnormal data removal, law vector calculate adjustment, absent region repair.
The invention discloses the grid surface method for reconstructing of the scene understanding using the system is also disclosed, specific steps are such as
Under:
Step 1: obtaining point cloud model using optics or non-optical equipment, and input to be originated from the point Yun Mo of different scenes
Type;
Step 2: carrying out disorder data recognition, law vector calculating and consistency adjustment to the data of input;The exception of point cloud
Data identification and normal vector calculate, comprising: disorder data recognition and law vector calculating based on non-directed graph network are carried out to cloud
The pretreatment of adjustment.
Step 3: carrying out the broad sense geometric graphic element region segmentation of vision perception characteristic to cloud;Based on " first extracting branch area
The point cloud segmentation thinking in domain, then be sliced remaining area " is extracted in the low dimensional of model presented with 2D curved surface or 1D curve in spy
Reference breath realizes corresponding " block " the body dividing method of stub area point cloud as the basic basis of visual perception segmentation;For point
Point cloud after branch extracted region examines analysis according to model, to remaining complex shape area after taking-up stub area
Domain or structuring significantly put cloud and use the division based on " sheet body " more reasonable.It adopts the visible area of view-based access control model projection mapping
Set method is unanimously uniting the analysis dimensionality reduction to three-dimensional point cloud to the analysis to two-dimensional depth figure based on vertex in law vector
One partitioning algorithm, realizing there is " piece " shape of local visual perception to divide complex shape region.To finally realize point cloud
Region segmentation with vision perception characteristic;
Step 4: to the point cloud broad sense geometric graphic element after segmentation carry out self study understanding based on scene and absent region from
It is dynamic to repair;The self study of scene understand be between segmentation result geometric shape and physical relationship explicitly marked
Will, classification extract the solid in segmentation result with close form.For object similar in form according to same characteristic features
Its mapping value of the region of type also repair automatically as a result, being converted into the image repair algorithm realization based on template matching by similar analysis
It is multiple.For the missing with bright stub area, reparation is realized automatically in such a way that section is scanned;For the structurings such as CAD point
Cloud can be automatically repaired by the way of element figure logical operation.
Step 5: carrying out the quasi- of vision hypersphere binary mapping to the cut-point cloud after carrying out scene understanding and being automatically repaired
Computational geometry point cloud surface is rebuild;The mapping of vision hypersphere millet cake cloud is that details (oxeye scrutinizes wisp), remote is seen according to low coverage
Away from the characteristic for seeing profile (pigsney slightly sees big object), can clearly be observed on object when eyes are relative to object infinity
Pole fine feature is much larger than the sphere for being observed dimension of object with radius R instead of eyes, and overturn to imaging of standing upside down, obtained
To the basic principle of visible field of view hypersphere mapping overturning.
Step 6: corresponding to the point cloud grid model after rebuilding, reconstructed results follow ρ-ε and rebuild criterion.
Further, the step two specifically:
2.1, settingIndicate the object of real scene,Indicate object boundary,It is rightIt carries out
Collected scattered data points set,Set is closed on for K of the point p in point set P, the non-directed graph G of building point cloud (V,
E), V=P, E are by the vertex P to the side collection constituted;Data pi,pjMeasuring similarity function be represented by k (pi,pj), pi,pj
Metric function should take the larger value when being subordinate to same subspace, otherwise smaller, qpi,qpjFor pi,pjCorresponding attribute mapping value;
2.2, the analysis to following objective function solving result can be converted into the analysis of cloud abnormal data:
The characteristic value of LX=λ X, feature vector computational problem are converted by the identification problem of abnormal data;
2.3, to there are the point cloud normal vectors of noise, with by extracting the inherent curved surface number of maximum that local neighborhood is contained
According to, and by its curved projection surfaces space to estimate the accurate normal vector of a cloud, and design the method that point cloud data keeps feature
To vector adjustment algorithm;
2.4, cluster subspace based on data local optimum constructs similar function and measures the similarity between normal direction,
The consistency adjustment algorithm of design data point cloud method arrow accordingly.
Further, the step three specifically:
3.1, complicated scene is divided into according to the feature of regional area corresponding by multiple basic " blocks " or " piece "
Broad sense geometric graphic element assembly is uniformly converted into a Problems of Reconstruction for cloud overall model the reconstruction to broad sense geometric graphic element;
3.2, be based on " first extracting branch region, then be sliced remaining area " point cloud segmentation thinking, extract model with 2D
The low dimensional internal characteristics information that curved surface or 1D curve are presented, the basic basis as visual perception segmentation;
3.3, the visible area acquisition method of view-based access control model projection mapping, by the analysis dimensionality reduction to three-dimensional point cloud to two dimension
The analysis of depth map, based on, in consistent unified partitioning algorithm, realizing has office to complex shape region in vertex and law vector
" piece " shape of portion's visual perception is divided, to finally realize that point cloud has the region segmentation of vision perception characteristic;
3.4, normal direction n institute characteristic on straight line is corresponded at it according to skeletal point, design is towards opening, noise surface points cloud
Approximate skeletal point derivation algorithm:
For a cloudCorresponding skeleton;(p, q) corresponding maximum inscribed sphereCentre of sphere cp=q=p- ρpnp,For BpWithPoint of contact, θpFor the centre of sphere to two point of contacts
Subtended angle, radiusThe solution of skeletal point can be converted into iterative calculationIt is
No the problem of being 1;
The measuring similarity function for containing vertex position, radius information etc. is represented by k (Bpi,Bpj)=f1(||pi-pj|
|)f2(|ρi-ρj|), using the optimization objective function of disorder data recognition, the geological information according to local form is recalculated
Accurate data information;
Skeletal point is represented byCorresponding 1D structural framework center curve is therefrom extracted to be expressed asThe extraction of frame center's curve is converted into the similar AX=BQ equation of solution, and A is strictly diagonally dominant matrices;
3.5, to the result that obtains of solution using the Moving Least Squares curve matching based on MLS and based on principal component analysis
The method of projection obtains final frame center's curve, and homologous thread branches into
Judgment criterion is scanned using frame center's curve and combination, extraction obtains visual perception apparent " minimum " and changes branch
Region, the judgment criterion for scanning foundation may be designed as:
g(θi| Θ) bigger, θiIt is more excellent, as g (θi| Θ) it is less than given threshold value time delay branch and scans stopping;
The vertical plane of skeleton curve branch is expressed as π⊥, θ is as vertical plane π⊥Close region,It is preceding
K adjacent close region extracts as a result, kernel function f (x) calculates θ afterwardsiSimilarity between closing on;
Skeletal point corresponds to inscribed sphere BpUnionModel Ω can accurately retouched based on body mode
It states, therefore the visibility of point p is converted to whether it falls in ballVisible range in, to be mapped using depth map
Mode realize the extraction to visual field visibility region point cloud;
3.6, construct depth value mapping function T of the unit ball on the square grid plate of the H × H parallel with view plane
(u, v), T (u, v) are that unit ball existsThe corresponding depth value in position;To allAfter carrying out rasterizing mapping
It obtains model and corresponds to depth map matrix D on visual field direction, the visibility of point p can be calculated directly by its depth value;From visual field
L*The visible point cloud being just drawn up is PV, according to D and PVBetween corresponding relationship, in D corresponding pixel points carry out plane grid
P can be obtained in changeVCorresponding space lattice MV;Entire point cloud model reflect based on depth map from multiple and different visual field directions
The segmentation penetrated finally obtains the segmentation result with visual perception to entire point cloud model.
Further, the step four is specially and is split result to point cloud model P to indicate are as follows:χiTo divide
That cuts out has the simple shape point cloud sector domain of basic " block ", " piece " shape, designs self study function, according to its topology, geometric form
State feature closes on relationship in physical space, these simple " blocks ", " piece " are sorted out, combined, realizes the reason to scene
Solution.
Further, the step four is specific as follows:
4.1, classification function be used under design method:
For group indication vector, YiIt is χiCorresponding group indication;K indicates shared K classification
Standard,Indicate χiWhether the probability of k-th classification, f are belonged toW(X, Y) reflects all segmentation results
Between maximum similarity;
4.2, non-directed graph G (V, E), the V={ 1 ..., N } of piecemeal X are constructed,χiIt is corresponding to classify for being used as
The maps feature vectors of foundation are φn(χi),For classified weight;φr(χi,χj) it is phase between two piecemeals of measurement
Like the feature vector of degree relationship, show whether it is under the jurisdiction of same or same type objects, respective weightsMeasurement two
Dependence between classification;The classification results of XBy having the discriminant function f with weight vectors WW(X, Y) is solved:
ΓOFor the jobbie being made of after categorized basic " block ", " piece ", Γ is the set of corresponding object.
Further, the step five specifically:
5.1, the basic principle of view-based access control model hypersphere millet cake cloud mapping indicates that mapping radius, C are point of observation, hypersphere with R
Mapping function fP(R, C) may be designed as:
After being computed, point cloud model visibility region is mapped as close in the ultra-thin spherical shell of spherical surface, non-visible cloud maps
For in spherical surface;Because of the hypersphere millet cake cloud topological homeomorphism after the source point cloud of visibility region and mapping, therefore by putting cloud institute structure after mapping
The corresponding triangle gridding expression of visibility region source point cloud can be obtained in the convex closure surface mesh built, and can be used for what hypersphere mapped
It realizes in the segmentation and mesh reconstruction of spatial point cloud;
5.2, the binary segmentation to model point cloud data is realized based on positive and negative binary segmentation;It is reflected using hypersphere millet cake cloud
The comformity relation and mapping accuracy constraint condition between convex closure and local visual angle visible point cloud reconstruction are penetrated, to positive and negative binary
The convex closure at visual angle rebuild grid carry out reflection penetrate, merge after the integral grid reconstruction model of corresponding pel point cloud can be realized;?
On the basis of carrying out data processing, segmentation, self study classification to cloud, in segmentation resultEach is simple wide
Adopted geometric graphic element χiIt merges, that is, is able to achieve to entire object using the reconstruction mapped based on binary visual angle, and between pel
The fine granularity geometrical reconstruction of body.
The beneficial effect of the present invention and the prior art is: the present invention is in the point cloud data reconstruction from different scenes
Complex scene divide based on the junior unit with local visual perception characteristics by existed general problem first, deeply divides
Space structure, the geometric shape information for analysing sub- cut zone, construct corresponding self study classification model, view-based access control model hypersphere
Mapping principle puts forward a whole set of data analysis and high-quality surface Model Reconstruction algorithm with underlying semantics characteristic, the calculation
Method has the complexity of O (NlogN), and geometric topology information of the reconstructed results independent of cloud, interpolation is in source input point cloud, completely
Foot uses ρ-ε when geometrical reconstruction to rebuild criterion.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the grid surface method for reconstructing of scene understanding of the present invention;
Fig. 2 is the structural schematic diagram that the present invention realizes the method for the present invention.
Specific embodiment
It is understandable to enable the above objects, features, and advantages of the embodiment of the present invention to become apparent, it is right with reference to the accompanying drawing
Specific embodiments of the present invention are described in detail.
The embodiment of the invention provides a kind of grid surface method for reconstructing of scene understanding, as shown in Figure 1, specific steps
Are as follows:
Step S101, to be originated from the point cloud model of different scenes.When obtaining point cloud model using optical device, according to light
Rectilinear propagation properties, when single measurement, can only obtain the unidirectional local perspective data of object.In measurement large scale scene or form
When the model of complexity, for the omnidirectional's model data for obtaining testee, it is necessary to be taken multiple measurements from multiple visual angles and splicing is melted
The problems such as closing, reducing uneven sampling, the miscellaneous point of noise, shortage of data from source.When obtaining point cloud model using non-optical equipment,
Such as: CT, MRI need to denoise image data, feature enhancing to obtain accurate 3-D data set, quasi- to reduce noise
Really extract effective information.
Step S102 carries out disorder data recognition, law vector calculating and consistency adjustment to the data of input.Point cloud data
Usually there is by multiple parts the subspace structure composition of certain feature, reappear these subspaces can accurately distinguish it is normal and
Abnormal data.
Indicate the object of real scene,Indicate object boundary,It is rightIt is acquired
Scattered data points set,Set is closed on for K of the point p in point set P.The non-directed graph G (V, E), V=P, E of building point cloud
For by the vertex P to the side collection constituted.
Data pi,pjMeasuring similarity function be represented by k (pi,pj), pi,pjMetric function when being subordinate to same subspace
The larger value should be taken, on the contrary smaller, qpi,qpjFor pi,pjCorresponding attribute mapping value.An analysis for cloud abnormal data can be converted into
Analysis to following objective function solving result:
By deriving, in embodiments of the present invention, the characteristic value of LX=λ X, spy are converted by the identification problem of abnormal data
Levy vector computational problem.By investigating the relationship between characteristic value, feature vector, corresponding judgement, criterion automatic identification are designed
Abnormal data.
To there are the point cloud normal vectors of noise, the inherent curved surface data of maximum that local neighborhood is contained is extracted with passing through,
And by its curved projection surfaces space to estimate the accurate normal vector of a cloud, and design the normal direction arrow that point cloud data keeps feature
Measure adjustment algorithm.
Tuning function design criteria is that the biggish adjacent region data weighing factor of law vector deviation should be smaller, and deviation is lesser
Adjacent region data then has larger impact.Cluster subspace based on data local optimum can construct similar function to measure normal direction
Between similarity, accordingly design data point cloud method arrow consistency adjustment algorithm.
Step S103 carries out the broad sense geometric graphic element region segmentation of vision perception characteristic to cloud.Point cloud model is corresponding
Scene be usually be made of multiple objects, between object and object, between object itself different parts usually in vision or space
There can be apparent boundary on position.
Complicated scene is divided into according to the feature of regional area by multiple basic " block " or " piece " corresponding broad sense
Geometric graphic element assembly is uniformly converted into a Problems of Reconstruction for cloud overall model the reconstruction to broad sense geometric graphic element.
Based on the point cloud segmentation thinking in " first extracting branch region, then be sliced remaining area ", extract model with 2D curved surface
Or the low dimensional internal characteristics information that 1D curve is presented realizes stub area point cloud as the basic basis of visual perception segmentation
Corresponding " block " body dividing method;Point cloud after extracting for stub area, examines analysis according to model, to taking-up
It is more reasonable using the division based on " sheet body " significantly to put cloud for remaining complex shape region or structuring after stub area.Base
In the visible area acquisition method of visual projection's mapping, by the analysis dimensionality reduction to three-dimensional point cloud to the analysis to two-dimensional depth figure,
Based on, in consistent unified partitioning algorithm, realizing has local visual perception to complex shape region in vertex and law vector
The segmentation of " piece " shape.To finally realize that point cloud has the region segmentation of vision perception characteristic.
Correspond to normal direction n institute characteristic on straight line at it according to skeletal point, design towards opening, noise surface points cloud it is close
Like skeletal point derivation algorithm.
For a cloudCorresponding skeleton.(p, q) corresponding maximum inscribed sphereCentre of sphere cp=q=p- ρpnp,For BpWithPoint of contact, θpFor the centre of sphere to two point of contacts
Subtended angle, radius
The solution of skeletal point can be converted into iterative calculationThe problem of whether being 1.Having
During body is implemented, the measuring similarity function for containing vertex position, radius information etc. is represented by k (Bpi,Bpj)=f1(||pi-pj|
|)f2(|ρi-ρj|), using the optimization objective function of disorder data recognition, the geological information according to local form is recalculated
Accurate data information.
Skeletal point is represented byCorresponding 1D structural framework center curve is therefrom extracted to be expressed asThe extraction of frame center's curve is converted into the similar AX=BQ equation of solution, and A is strictly diagonally dominant matrices.To solution
The result obtained is obtained finally using the Moving Least Squares curve matching based on MLS and the method based on principal component analysis projection
Frame center's curve, homologous thread branches into
Judgment criterion is scanned using frame center's curve and combination, extraction obtains visual perception apparent " minimum " and changes branch
Region.The judgment criterion for scanning foundation may be designed as:
g(θi| Θ) bigger, θiIt is more excellent, as g (θi| Θ) it is less than given threshold value time delay branch and scans stopping.
The vertical plane of skeleton curve branch is expressed as π⊥, θ is as vertical plane π⊥Close region,It is preceding
K adjacent close region extracts as a result, kernel function f (x) calculates θ afterwardsiSimilarity between closing on.
Simulated optical method obtain body surface data, be from multiple views to data splicing fusion obtained by
Principle, the method that view-based access control model projection mapping can be used in point cloud data realizes the reversed extraction segmentation of visible area, and dimensionality reduction reflects
It is mapped to two-dimensional image space, realizes and the more efficiently analysis based on grid is carried out to the point cloud sector domain of complex shape.
Skeletal point corresponds to inscribed sphere BpUnionModel Ω can accurately retouched based on body mode
It states, therefore the visibility of point p is converted to whether it falls in ballVisible range in, so as to be reflected using depth map
The mode penetrated realizes the extraction to visual field visibility region point cloud.
Construct unit ball on the square grid plate of the H × H parallel with view plane depth value mapping function T (u,
V), T (u, v) is that unit ball existsThe corresponding depth value in position.To allIt can be obtained after carrying out rasterizing mapping
Model corresponds to the depth map matrix D on visual field direction, and the visibility of point p can be calculated directly by its depth value.From visual field L*Side
The visible point cloud being drawn up is PV, according to D and PVBetween corresponding relationship, in D corresponding pixel points carry out plane net format i.e.
P can be obtainedVCorresponding space lattice MV.Entire point cloud model is carried out from multiple and different visual field directions based on depth map mapping
Segmentation, finally obtains the segmentation result with visual perception to entire point cloud model.
Step S104 carries out self study understanding and absent region based on scene to the point cloud broad sense geometric graphic element after segmentation
It is automatically repaired.The self study of scene understand be between segmentation result geometric shape and physical relationship explicitly marked
Will, classification extract the solid in segmentation result with close form.For object similar in form according to same characteristic features
Its mapping value of the region of type also repair automatically as a result, being converted into the image repair algorithm realization based on template matching by similar analysis
It is multiple.For the missing with bright stub area, reparation is realized automatically in such a way that section is scanned;For the structurings such as CAD point
Cloud can be automatically repaired by the way of element figure logical operation.
Being split result to point cloud model P may be expressed as:χiThere is basic " block ", " piece " for what is be partitioned into
Self study function is designed in the simple shape point cloud sector domain of shape, according to its topology, geometrical feature, closing in physical space
These simple " blocks ", " piece " are sorted out, are combined, realize the understanding to scene by relationship.
Classification function can be used under design method:
For group indication vector, YiIt is χiCorresponding group indication.K indicates shared K classification
Standard,Indicate χiWhether the probability of k-th classification, f are belonged toW(X, Y) reflects all segmentation results
Between maximum similarity.Non-directed graph G (V, E), the V={ 1 ..., N } of piecemeal X are constructed,χiIt is corresponding to be used to make
Maps feature vectors for classification foundation are φn(χi),For classified weight.φr(χi,χj) it is two piecemeals of measurement
Between similarity relationship feature vector, show whether it is under the jurisdiction of same or same type objects, respective weights
Dependence between two classification of measurement.The classification results of XIt can be by having the discriminant function f with weight vectors WW(X,
Y it) solves:
ΓOFor the jobbie being made of after categorized basic " block ", " piece ", Γ is the set of corresponding object.φn(χi)
Can the feature according to corresponding to the different submodel such as industry pattern, CAD model, buildings model, organism be designed.Such as: needle
To CAD model φn(χi) can be corresponded to parameter by the plane for reflecting basic geometric graphic element, ball, cylinder, circular cone, annulus and be formed,
The sculpture model of complicated form can use the corresponding characteristic value of principal component analysis, the feature vector etc. for reflecting cut zone geometric shape
Information composition.
Step S105 carries out the mapping of vision hypersphere binary to the cut-point cloud after carrying out scene understanding and being automatically repaired
Quasi- computational geometry point cloud surface is rebuild.Vision hypersphere millet cake cloud mapping be seen according to low coverage details (oxeye scrutinizes wisp),
Long distance sees the characteristic of profile (pigsney slightly sees big object), can clearly observe on object when eyes are relative to object infinity
Pole fine feature, with radius R be much larger than be observed dimension of object sphere replace eyes, and to stand upside down imaging overturn,
Obtain the basic principle of visible field of view hypersphere mapping overturning.Indicate that mapping radius, C are point of observation, hypersphere mapping with R
Function fP(R, C) may be designed as:
After above-mentioned formula calculates, point cloud model visibility region is mapped as close in the ultra-thin spherical shell of spherical surface (spherical shell thickness
Spend δ < < R, abbreviation hypersphere), non-visible cloud be mapped as in spherical surface.
Because of the hypersphere millet cake cloud topological homeomorphism after the source point cloud of visibility region and mapping, therefore as being put constructed by cloud after mapping
Convex closure surface mesh the corresponding triangle gridding expression of visibility region source point cloud can be obtained, reality can be used for by what hypersphere mapped
Now in the segmentation and mesh reconstruction of spatial point cloud.
Positive and negative binary segmentation is when observing simple objects according to the mankind, to be seen twice from positive and negative binary visual angle to object
It examines, can substantially see the overall picture of object, realize the binary segmentation to model point cloud data based on this.Utilize hypersphere millet cake
Cloud maps comformity relation and mapping accuracy constraint condition between convex closure and local visual angle visible point cloud reconstruction, to positive and negative
The convex closure at binary visual angle rebuild grid carry out reflection penetrate, merge after the integral grid reconstruction mould of corresponding pel point cloud can be realized
Type.
It can first construct towards Octree deblocking and kdtree neighborhood search data structure based on GPU+CPU, right
On the basis of point cloud carries out data processing, segmentation, self study classification, in segmentation resultEach simple broad sense
Geometric graphic element χiIt merges, that is, is able to achieve to whole object using the reconstruction mapped based on binary visual angle, and between pel
Fine granularity geometrical reconstruction.
Step S106 corresponds to corresponding cloud grid model after rebuilding.When parameter selection is reasonable, reconstructed results follow ρ-
ε rebuilds criterion.
The above method is directed to be originated from the point cloud data of different scenes, the spy according to point cloud model therein composed structure
Property, the point cloud model dividing method with vision perception characteristic is designed, the Problems of Reconstruction of complex scene model is subjected to dimensionality reduction;With
The self study framework that segmentation result understands as basic unit building model scene, is realized to absent region according to scene classification
It is automatically repaired;Simulation and abstract visual imaging theory, realization follow ρ-ε and rebuild criterion, the grid weight independent of sampling density
Build algorithm.
As shown in Fig. 2, the invention also discloses a kind of grid surface reconstructing systems of scene understanding, comprising: data input
Unit 201, data processing unit 202, curve reestablishing unit 207 put cloud grid model output unit 210.Wherein, data processing
Unit includes: point cloud pretreatment unit 203, point cloud segmentation unit 204, point cloud reparation unit 205;Curve reestablishing unit 207 wraps
It includes: hypersphere map unit 208, binary segmentation reconstruction unit 209.
Data input cell 201, for reading and parsing from different scenes, different types of cloud 3-D data set.
Data processing unit 202, the segmentation for carrying out that there is vision perception characteristic to cloud, and to local shortage of data domain area into
Row being automatically repaired based on scene understanding.Point cloud pretreatment unit 203 is calculated for the disorder data recognition to cloud, removal
It obtains towards consistent cloud law vector.Point cloud segmentation unit 204, for carrying out point cloud model based on " block ", " sheet body "
Segmentation, to obtain broad sense geometric graphic element collective data collection used for reconstruction.Point cloud repairs unit 205, is used for broad sense geometry
Pel carries out being automatically repaired for the localized loss data of scene self study understanding.Pretreated point cloud data unit 206, is used for
Input data as curve reestablishing unit 207.Curve reestablishing unit 207, for the number exported via data processing unit
According to rebuild based on the positive and negative visual angle binary segmentation algorithm for reconstructing of mapping " hypersphere " grid surface.
Hypersphere map unit 208, for converting hypersphere mapping point for the reconstruction of point cloud local visual angle visibility region
The convex closure of cloud solves.Binary segmentation reconstruction unit 209, for realizing to broad sense geometric graphic element based on positive and negative binary visual angle
Convex closure is rebuild.Point cloud grid model output unit 210, for exporting the grid model after rebuilding.Data input cell 201, institute
The data format of receiving includes: asc, vtx, pcd, pts etc., and source includes: optics (structure light, laser, LIDAR etc.), non-light
It learns scanner (CT, MRI).The data of input usually have data volume big (vertex for including mostly is counted with ten million, hundred million), exist and make an uproar
The characteristics of sound and abnormal data, therefore usually require to be handled by data processing unit 202.
The main operational steps of data processing are point cloud pretreatment, point cloud segmentation, point Yun Xiufu, and it is described for respectively corresponding
203, Unit 204,205.Wherein described 203,205 operating units are optional.Such as: if the point cloud data of input is soft via other
What part or tool be treated, comprising: abnormal data removal, law vector calculate adjustment, absent region is repaired, then described
203, the operation of Unit 205 can omit.In special circumstances, described if the point cloud data of input is simple basic geometric graphic element
Unit 204 can omit.
In specific implementation, the point cloud pretreatment unit 203, carries out identification removal to abnormal data first, then into
Row law vector calculates, and finally carries out consistency adjustment.
In specific implementation, using optimization objective function:By abnormal data
Identification problem is converted into the characteristic value of LX=λ X, feature vector computational problem.By investigating the pass between characteristic value, feature vector
System, accurate automatic identification abnormal data.Wherein, k (pi,pj) it is data pi,pjMeasuring similarity function, qpi,qpjFor pi,pj
Corresponding attribute mapping value.
To reduce the influence that noise calculates law vector, using the side for extracting the inherent curved surface of maximum that local neighborhood is contained
Method improves the accuracy that law vector calculates, and constructs law vector similarity function, carries out normal vector consistency adjustment.
After carrying out noise remove, normal direction information calculating to cloud, 204 couples of point Yun Jinhang of the point cloud segmentation unit are utilized
The segmentation of broad sense geometric graphic element.
" block " shape area preference apparent to branch extracts segmentation, then cuts " piece " to remaining area and extracts segmentation." block "
Body segmentation is depended on dependent on the low dimensional internal characteristics information indicated with 2D curved surface or 1D curve, the segmentation of " piece " shape by three-dimensional point
Two-dimensional depth figure after cloud dimensionality reduction.
In such a way that frame center's curve and combination scan judgment criterion, extracts segmentation visual perception and significantly minimize
" block " body stub area.The solution that frame center's curve corresponds to skeletal point can be converted into iterative calculationThe problem of whether being 1.Wherein, θpSubtended angle for the maximum inscribed sphere centre of sphere to two point of contacts, ρp
For radius.
In specific implementation, the discriminant function of foundation is scanned are as follows:Wherein, θ is as vertical
Plane π⊥Close region,Result is extracted for K close region.As g (θi| Θ) it is less than given threshold value time delay branch
Scan stopping.
" piece " the shape extraction to visual field visibility region point cloud is realized by the way of depth map mapping.Entire point cloud model from
Multiple and different visual field directions carry out the segmentation mapped based on depth map, and finally obtaining has visual perception to entire point cloud model
Segmentation result.In segmentation, visual field direction is usually that positive and negative visual angle occurs in pairs, to reduce " piece " number after segmentation.
Unit 205 is managed in described cloud reparation, to the point cloud after segmentationFirst to χiIt carries out based on scene understanding
Classify, then to χiLocalized loss region carry out being automatically repaired based on scene understanding.Wherein, χiFor broad sense geometric graph
Member.
Classification function uses:Wherein,For group indication vector, YiIt is χiIt is corresponding
Group indication.According to classification results, image repair algorithm based on template matching is respectively adopted for localized loss region, cuts
The algorithm of algorithm, element figure logical operation that face is scanned is automatically repaired.Via data processing unit 202 (comprising optional
Operating unit: 203, output data 204,205) after operation processing corresponds to Unit 206, include that the necessary geometry of model is believed
Breath, such as: law vector, and data after localized loss it is repaired after tended to be complete, to realize accurate Model Reconstruction.
The pretreated point cloud data unit 206, the input data as curve reestablishing unit 207.The curved surface
Reconstruction unit 207 first reflects visibility region point cloud using the basic principle of visible field of view hypersphere mapping overturning
It penetrates, according to the hypersphere millet cake cloud topological homeomorphism after the source point cloud of visibility region and mapping, realizes the reconstruction to local visual angle grid;
Then reconstruction grid of the positive and negative binary visual angle after hypersphere maps is merged, realizes the whole of corresponding broad sense pel point cloud
Volume mesh reconstruction model;Finally in segmentation resultEach simple broad sense geometric graphic element χiUsing based on binary
The reconstruction of visual angle mapping, and merged between pel, it realizes to entire point cloud model curve reestablishing.Hypersphere mapping function
fP(R, C) is used:Wherein, R indicates that mapping radius, C are to see
It examines a little.R is after rebuilding a large amount of models, empirically, can extrapolate an optimal value.
When positive and negative binary visual angle is divided, the principle that visual angle selects includes that number of vertices is up to preferred for partial visible point cloud
Direction.Fusion inside broad sense geometric graphic element, the fusion between pel, can uniformly be handled.
In specific implementation, to make system that there is preferable efficiency, real-time, using being based on GPU+ when Data Structure Design
The MIDAS mixed data structure of the Octree of CPU, kdtree.
Those of ordinary skill in the art will appreciate that all or part of the steps in the various methods of above-described embodiment is can
It is completed with instructing relevant hardware by program, which can be stored in a computer readable storage medium, storage
Medium may include: ROM, RAM, disk or CD etc..
Although present disclosure is as above, present invention is not limited to this.Anyone skilled in the art are not departing from this
It in the spirit and scope of invention, can make various changes or modifications, therefore protection scope of the present invention should be with claim institute
Subject to the range of restriction.
Claims (5)
1. a kind of grid surface reconstructing system of scene understanding, which is characterized in that the system includes data input cell, number
According to processing unit, curve reestablishing unit, point cloud network surface model output unit;The data processing unit includes that point cloud is pre-
Processing unit, point cloud segmentation unit, point cloud repair unit;The curve reestablishing unit includes hypersphere map unit, binary
Divide reconstruction unit.
2. a kind of grid surface reconstructing system of scene understanding according to claim 1, which is characterized in that the data
The data that input unit is received are the 3-D data sets of a cloud, and format includes asc, vtx, pcd, pts, source include optics,
Non-optical scanner.
3. a kind of grid surface reconstructing system of scene understanding according to claim 1, which is characterized in that the data
Input unit is used to input the 3-D data set to curve reestablishing point cloud;The data processing unit be used for input point cloud into
Row point cloud pretreatment, point cloud segmentation, point Yun Xiufu;The point cloud pretreatment unit is used to calculate a cloud denoising, law vector
And consistency adjustment;The point cloud segmentation unit is used to carry out a cloud segmentation of block, sheet body;The point cloud is repaired single
Member is for being automatically repaired cloud progress scene self study understanding, the classification of missing data;The curve reestablishing unit is used for
The increment of the whole body reconstruction, neighbor dependency that rely on the vision that cloud mapped based on hypersphere is rebuild;The hypersphere
The convex closure that map unit is used to convert the reconstruction of point cloud local visual angle visibility region to hypersphere mapping point cloud solves;Described
Binary segmentation reconstruction unit is rebuild for realizing the convex closure based on positive and negative binary visual angle to broad sense geometric graphic element;The point
Cloud grid model output unit is used to export the grid model after rebuilding.
4. a kind of grid surface reconstructing system of scene understanding according to claim 1, which is characterized in that the data
The operating procedure of processing be point cloud pretreatment, point cloud segmentation, point Yun Xiufu, respectively correspond for the point cloud pretreatment unit,
Point cloud segmentation unit, point cloud repair unit;The point cloud pretreatment unit, that point cloud repairs unit is optional;It is specific as follows:
If the point cloud data of input has been treated via other software or tool, the point cloud pretreatment list
The operation that member, point cloud repair unit is omitted;
If the point cloud data of input is simple basic geometric graphic element, the point cloud segmentation unit can be omitted.
5. a kind of grid surface reconstructing system of scene understanding according to claim 4, which is characterized in that the input
Point cloud data be to be treated via other software or tool, it is therein processing include: abnormal data removal, method
Vector calculates adjustment, absent region is repaired.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910131449.4A CN110009671B (en) | 2019-02-22 | 2019-02-22 | Grid curved surface reconstruction system for scene understanding |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910131449.4A CN110009671B (en) | 2019-02-22 | 2019-02-22 | Grid curved surface reconstruction system for scene understanding |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110009671A true CN110009671A (en) | 2019-07-12 |
CN110009671B CN110009671B (en) | 2023-06-09 |
Family
ID=67165904
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910131449.4A Active CN110009671B (en) | 2019-02-22 | 2019-02-22 | Grid curved surface reconstruction system for scene understanding |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110009671B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112288331A (en) * | 2020-11-24 | 2021-01-29 | 哈尔滨工业大学 | Building reconstruction evaluation system based on semantic recognition |
CN112686916A (en) * | 2020-12-28 | 2021-04-20 | 淮阴工学院 | Curved surface reconstruction system based on heterogeneous multi-region CT scanning data processing |
CN113591556A (en) * | 2021-06-22 | 2021-11-02 | 长春理工大学 | Three-dimensional point cloud semantic analysis method based on neural network three-body model |
CN117036634A (en) * | 2023-10-08 | 2023-11-10 | 青岛星邦光电科技有限责任公司 | Automatic construction method for three-dimensional scene of smart city |
WO2023220977A1 (en) * | 2022-05-18 | 2023-11-23 | Huawei Technologies Co., Ltd. | Method and device for detecting data |
CN112686909B (en) * | 2020-12-28 | 2024-04-19 | 淮阴工学院 | Heterogeneous multi-region CT scanning data processing-based multiphase implicit curved surface reconstruction method |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101938668A (en) * | 2010-09-10 | 2011-01-05 | 中国科学院自动化研究所 | Method for three-dimensional reconstruction of multilevel lens multi-view scene |
CN107016725A (en) * | 2017-02-27 | 2017-08-04 | 电子科技大学 | A kind of vegetation three-dimensional live modeling method for taking LiDAR point cloud data distribution difference into account |
CN108088386A (en) * | 2017-12-15 | 2018-05-29 | 深圳大学 | The binary area-structure light detection method and system of a kind of micro-nano magnitude |
-
2019
- 2019-02-22 CN CN201910131449.4A patent/CN110009671B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101938668A (en) * | 2010-09-10 | 2011-01-05 | 中国科学院自动化研究所 | Method for three-dimensional reconstruction of multilevel lens multi-view scene |
CN107016725A (en) * | 2017-02-27 | 2017-08-04 | 电子科技大学 | A kind of vegetation three-dimensional live modeling method for taking LiDAR point cloud data distribution difference into account |
CN108088386A (en) * | 2017-12-15 | 2018-05-29 | 深圳大学 | The binary area-structure light detection method and system of a kind of micro-nano magnitude |
Non-Patent Citations (1)
Title |
---|
孙志敬: "面向船体外板加工成型过程的三维点云重建技术研究与实现", 《中国优秀硕士论文全文数据库》 * |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112288331A (en) * | 2020-11-24 | 2021-01-29 | 哈尔滨工业大学 | Building reconstruction evaluation system based on semantic recognition |
CN112288331B (en) * | 2020-11-24 | 2022-04-12 | 哈尔滨工业大学 | Building reconstruction evaluation system based on semantic recognition |
CN112686916A (en) * | 2020-12-28 | 2021-04-20 | 淮阴工学院 | Curved surface reconstruction system based on heterogeneous multi-region CT scanning data processing |
CN112686916B (en) * | 2020-12-28 | 2024-04-05 | 淮阴工学院 | Curved surface reconstruction system based on heterogeneous multi-region CT scanning data processing |
CN112686909B (en) * | 2020-12-28 | 2024-04-19 | 淮阴工学院 | Heterogeneous multi-region CT scanning data processing-based multiphase implicit curved surface reconstruction method |
CN113591556A (en) * | 2021-06-22 | 2021-11-02 | 长春理工大学 | Three-dimensional point cloud semantic analysis method based on neural network three-body model |
WO2023220977A1 (en) * | 2022-05-18 | 2023-11-23 | Huawei Technologies Co., Ltd. | Method and device for detecting data |
CN117036634A (en) * | 2023-10-08 | 2023-11-10 | 青岛星邦光电科技有限责任公司 | Automatic construction method for three-dimensional scene of smart city |
Also Published As
Publication number | Publication date |
---|---|
CN110009671B (en) | 2023-06-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110009743A (en) | A kind of grid surface method for reconstructing of scene understanding | |
CN110009671A (en) | A kind of grid surface reconstructing system of scene understanding | |
Gibbs et al. | Approaches to three-dimensional reconstruction of plant shoot topology and geometry | |
CN110009727A (en) | A kind of indoor threedimensional model automatic reconfiguration method and system with structure semantics | |
CN108090960A (en) | A kind of Object reconstruction method based on geometrical constraint | |
CN102222357B (en) | Foot-shaped three-dimensional surface reconstruction method based on image segmentation and grid subdivision | |
CN108257213B (en) | Point cloud lightweight polygonal curved surface reconstruction method | |
CN109191510B (en) | 3D reconstruction method and device for pathological section | |
Pound et al. | A patch-based approach to 3D plant shoot phenotyping | |
CN104616349A (en) | Local curved surface change factor based scattered point cloud data compaction processing method | |
CN111783798B (en) | Mask generation method for simulated residual point cloud based on significance characteristics | |
Song et al. | Volumetric stereo and silhouette fusion for image-based modeling | |
CN113763529B (en) | Substation modeling method based on three-dimensional scanning | |
Haala et al. | Cell decomposition for the generation of building models at multiple scales | |
US8948498B1 (en) | Systems and methods to transform a colored point cloud to a 3D textured mesh | |
Pound et al. | Surface reconstruction of plant shoots from multiple views | |
CN107643048B (en) | Automatic extraction method of tree measuring factor based on point cloud data | |
CN115937465A (en) | Depth map fusion algorithm considering pixel region prediction | |
Wan et al. | Plot-level wood-leaf separation of trees using terrestrial LiDAR data based on a seg‐mentwise geometric feature classification method | |
Sahebdivani et al. | Deep learning based classification of color point cloud for 3D reconstruction of interior elements of buildings | |
Shi et al. | An integration method for scanned multi-view range images (MRIs) based on local weighted least squares (LWLS) surface fitting | |
CN110009744A (en) | A kind of the point cloud method for reconstructing and its system of vision mapping | |
Zou et al. | Research on optimization of rendering efficiency of point cloud data of transmission lines in three-dimensional GIS | |
Mariosa et al. | Surface Reconstruction from Range Images. | |
Wuttke et al. | Quality preserving fusion of 3d triangle meshes |
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 |