CN106709883A - Point cloud denoising method based on joint bilateral filtering and sharp feature skeleton extraction - Google Patents
Point cloud denoising method based on joint bilateral filtering and sharp feature skeleton extraction Download PDFInfo
- Publication number
- CN106709883A CN106709883A CN201611180960.6A CN201611180960A CN106709883A CN 106709883 A CN106709883 A CN 106709883A CN 201611180960 A CN201611180960 A CN 201611180960A CN 106709883 A CN106709883 A CN 106709883A
- Authority
- CN
- China
- Prior art keywords
- point
- normal vector
- bilateral filtering
- skeleton
- point 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
- 238000000034 method Methods 0.000 title claims abstract description 57
- 238000001914 filtration Methods 0.000 title claims abstract description 40
- 230000002146 bilateral effect Effects 0.000 title claims abstract description 35
- 238000000605 extraction Methods 0.000 title claims abstract description 14
- 239000013598 vector Substances 0.000 claims abstract description 109
- 238000012952 Resampling Methods 0.000 claims abstract description 11
- 238000005303 weighing Methods 0.000 claims description 11
- 238000005457 optimization Methods 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 3
- 241000792859 Enema Species 0.000 claims 1
- 230000013872 defecation Effects 0.000 claims 1
- 239000007920 enema Substances 0.000 claims 1
- 229940095399 enema Drugs 0.000 claims 1
- 230000001939 inductive effect Effects 0.000 claims 1
- 239000000829 suppository Substances 0.000 claims 1
- 238000004458 analytical method Methods 0.000 abstract description 5
- 230000000694 effects Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000010146 3D printing Methods 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 2
- 210000000988 bone and bone Anatomy 0.000 description 2
- 230000000052 comparative effect Effects 0.000 description 2
- 230000001737 promoting effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 239000010959 steel Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000003086 colorant Substances 0.000 description 1
- 238000004040 coloring Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 238000012916 structural analysis Methods 0.000 description 1
- 238000010408 sweeping Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20028—Bilateral filtering
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a point cloud denoising method based on joint bilateral filtering and sharp feature skeleton extraction, which comprises the steps of 1) acquiring candidate feature points by using a normal vector difference method; 2) extracting a skeleton of the candidate points; 3) performing resampling on the candidate points by using the skeleton so as to acquire feature points; 4) endowing each feature point with a multi-normal vector; filtering a point cloud normal vector based on joint bilateral filtering framework; and 6) updating the point position based on the filtered normal vector to acquire a denoised point cloud model. According to the point cloud denoising method, a sharp feature skeleton is firstly extracted to perform analysis on the model structure, so that sharp features of an object are well maintained while denoising is performed, thereby being conducive to dealing with high-intensity noise. The point cloud denoising method has the characteristic of high robustness and has excellent popularization and application prospects.
Description
Technical field
The present invention relates to computer graphics and three-dimensional point cloud denoising field, refer in particular to a kind of based on joint bilateral filtering
With the point cloud denoising method of sharp features skeletal extraction.
Background technology
The three-dimensional modeling data that most three-dimension sensor shoots is preserved with the form of a cloud.Because scanning sets
The limitation of standby precision, such as Microsoft Kinect, the cloud data for photographing inevitably all is made an uproar with certain
Sound.Therefore, point cloud denoising plays a very basic and important role in three-dimensional geometry process field.And put cloud denoising
Significant challenge is how to retain the minutia of object, the particularly holding of sharp features while denoising.
Joint bilateral filtering, as an extension of bilateral filtering technology, in image processing field it is verified that being one
Plant the effective method for keeping feature.Application of this technology in mesh denoising field is also achieved successfully.And this kind of filtering skill
The disposal ability of art, depends greatly on the structural information of image or grid, including aiming field to be determined.In digitized map
As in, sharp edges can be defined as the very big adjacent pixel of colouring discrimination, and different color regions can be completely in picture
Plain aspect makes a distinction.Therefore, pixel value can be described the marginal texture of image very well as the value of aiming field.In grid neck
Domain, each tri patch also can be to belong to certain face completely, therefore can very simply by the normal vector structure of tri patch
Build up aiming field.But in a cloud field, we do not have an element similar to tri patch, can uniquely belong to
Some face, certain point is likely located in the intersection in several faces, while belonging to multiple faces.Therefore, bilateral filtering side will be combined
It is very difficult that method is applied to a cloud denoising.
But pass through we have discovered that, the denoising of signature analysis addition point cloud can effectively be solved into this problem.Entering
Before row denoising, first according to the quantity and feature of each adjacent block put, classified.For different types of point, structure
Different aiming fields are built, and is added in joint bilateral filtering formula, be calculated filtered normal direction.And how to carry out spy
Levy analysis, it is necessary to which the feature skeleton to object is extracted, to help the overall structure of our preferably object analysis, more preferably
Ground retains sharp features while denoising.
The content of the invention
Shortcoming and defect it is an object of the invention to overcome prior art, there is provided a kind of effective, scientific and reasonable
Point cloud denoising method based on joint bilateral filtering and sharp features skeletal extraction.
To achieve the above object, technical scheme provided by the present invention is:Based on joint bilateral filtering and sharp features bone
The point cloud denoising method that frame is extracted, comprises the following steps:
1) candidate feature point is obtained using normal vector difference method;
2) to candidate point extracted region skeleton;
3) resampling is carried out to candidate point using skeleton, obtains characteristic point;
4) many normal vectors are assigned to characteristic point;
5) a cloud normal vector is filtered based on joint bilateral filtering framework;
6) position is updated based on filtered normal vector, obtains the point cloud model after denoising.
In step 1) in, candidate feature point is obtained using normal vector difference method, it is to calculate any two in each vertex neighborhood
The normal vector difference of point is simultaneously sued for peace, and when this summing value is more than predetermined threshold value, this point is just labeled as candidate feature point.
In step 2) in, it is to use l to candidate point extracted region skeleton1Intermediate value skeleton technique, to candidate feature point region
Calculated, obtained skeleton.
In step 3) in, resampling is carried out to candidate point using skeleton, obtain characteristic point, it is according to step 2) bone that obtains
Skeletal point on frame and skeleton, marks off a zonule, to the time fallen in the region in the middle of two neighboring skeletal point
Characteristic point is selected, one is selected, as characteristic point.
In step 4) in, many normal vectors are assigned to characteristic point, be according to step 3) judge the normal vector that the point should possess
Quantity, will be that k Neighbor Points carry out K-means clusters near k Neighbor Points of the point, and the normal vector of the point in each class is asked
Average value, obtains many normal vectors of the point.
In step 5) in, a cloud normal vector is filtered, it is necessary to first build aiming field based on joint bilateral filtering framework
Value, will guide thresholding to be set to guiding normal vector a little, each point and its k Neighbor Points is set into a block, by the block comprising the point
All include and;Each block is calculated its average normal vector a little, while calculating the vector difference of any two points in block
Average value, is defined as the continuity value of each block, and for weighing the vectorial uniformity of its internal point, continuity value is got over
Greatly, it was demonstrated that the block is across sharp features, and the point comprising another face is more;In order to retain sharp features, sharp spy is distinguished
The Different Plane of both sides is levied, it is necessary to what is selected is the minimum block of continuity value, i.e., the block of sharp features is not spanned across as far as possible, by this
The average normal vector of individual block, as the guiding normal vector of the point;
Joint bilateral filtering formula is as follows:
Wherein,It is piJ-th filtered normal vector of point, pk∈N(pi) show pkPoint is belonging to piNeighborhood of a point point,
nklIt is pkNormal vector before l-th filtering of point,It is alternate position spike weight, works as pkPoint with
piWhen the Euclidean distance of point is more remote, weight is smaller;It is normal vector difference weight, works as pkPoint
Guiding normal vector gklWith piThe guiding normal vector g of pointijNormal vector difference it is bigger when, weight is smaller;WijIt is normalization factor,
Ensure filtered normal vector mould a length of 1;
Work as pkPoint is when having multiple normal vectors, just select in its multiple normal vector with piThe point immediate normal vector of former normal vector
nkl, substitute into joint bilateral filtering formula and calculated, obtain piThe filtered normal vector of point.
In step 6) in, based on filtered normal vectorI represents the individual of model midpoint
Number, has n point;J represents the vectorial number that each point has, when i is general point, diWhen=1, i point are characterized, di> 1,
A position is updated, the point cloud model after denoising is obtained, it is necessary to the optimization problem being defined as follows:
εpointupd=εposition+εnormal (2)
Wherein:
In formula, piRepresent the point position, p ' after updatingiRepresent piPosition before renewal, NiRepresent p 'iK neighbour's point sets,
p′kRepresent p 'iK Neighbor Points, w (i, k)=wp(i,k)wn(i, k), whereinIt is used for
Position weighing factor of the neighborhood point to central point is weighed, when a distance is bigger, weight is smaller;It is used for
Normal vector weighing factor of the neighborhood point to central point is weighed, when the normal vector of the neighborhood point differs bigger with the normal vector of central point
When, weight is smaller;The iterative formula that solution formula (2) is obtained is:
Wherein t and t+1 is the t times and the t+1 times iteration, thus obtains new point positionBy iteration it is multiple after
Point cloud model after to denoising.
The present invention compared with prior art, has the following advantages that and beneficial effect:
1st, large scale noise is preferably processed.For existing for bilateral filtering and its related expanding method in denoising simultaneously
The problem of object sharp features can be obscured, the present invention is in the l using normal direction difference method binding characteristic region1Intermediate value skeleton is carried
Take, structural analysis integrally carried out to point cloud model, can before denoising sharp features present in first judgment models, going
Can accomplish targetedly to protect during making an uproar.Therefore, when we process the point cloud model of big noise, just can be in denoising
While, sharp features are kept well.As shown in fig. 6, the present invention is in the model for large scale noise, than other methods
There can be more preferable denoising effect, and its sharp features can be kept.
2nd, more preferable robustness.Although in the case where bilateral filtering framework is combined, it is necessary to set some relevant parameters, at this
In invention, the setting of relevant parameter is established and is correspondingly arranged by we with the size of model, is normalized for model size
Work, and to l1The generation of intermediate value skeleton has been also carried out optimization, therefore for most point cloud models, can directly transport
OK, and the excessive setting and adjustment of user is not needed.Meanwhile, the present invention can not only tackle all kinds of manikins, and for reality
The data of border scanning, the model data that such as Microsoft Kinect scannings are obtained, it is also possible to process well.
3rd, more easy usability.Whole process of the present invention is to carry out automatically, required parameter by algorithm according to size from
It is dynamic to calculate, and do not need user to interact setting, so for the user for using also without the Knowledge Base for having correlation,
Operation program is needed, required result can be just directly obtained.Therefore the present invention can also be conducive in China from now in 3D printing skill
In the popularization of art, the optimization for model quality provides help, therefore the present invention has very big actual promotional value.
Brief description of the drawings
Fig. 1 is the inventive method flow chart.
Fig. 2 a utilize l for the present invention1Intermediate value skeleton carries out the schematic diagram of resampling to edge point.
Fig. 2 b utilize l for the present invention1Intermediate value skeleton angle steel joint carries out the schematic diagram of resampling.
When Fig. 3 to a cloud normal vector based on joint bilateral filtering framework in the present invention to be filtered, the structure of aiming field
Schematic diagram.
Fig. 4 is the result after this denoising method treatment manikin, and is compared with other methods, wherein the method for the present invention
It is Ours.
Fig. 5 is the result after this denoising method treatment realistic model, and is compared with other methods, wherein the method for the present invention
It is Ours.
Fig. 6 is result of the present invention in model of the treatment with different scale noise, and is compared with other methods, wherein together
One is classified as with a method, and the method for the present invention is Ours;The same Noise Criterion of same behavior.Gaussian noise is respectively from top to bottom
1%, 2%, 3% and 4%.
Fig. 7 a are the design sketch that the present invention obtains candidate feature point using normal vector difference method.
Fig. 7 b are design sketch of the present invention to candidate point extracted region skeleton.
Fig. 7 c carry out resampling using skeleton for the present invention to candidate point, obtain the design sketch of characteristic point.
Fig. 8 is the design sketch that the present invention assigns many normal vectors to characteristic point.
Specific embodiment
With reference to specific embodiment, the invention will be further described.
As shown in figure 1, the point cloud denoising based on joint bilateral filtering and sharp features skeletal extraction that the present embodiment is provided
Method, comprises the following steps:
1) candidate feature point is obtained using normal vector difference method;
Calculate the normal vector difference of any two points in each vertex neighborhood and sue for peace, when this summing value is more than certain threshold value
When, this point is just labeled as candidate feature point, as shown in Figure 7a.
2) to candidate point extracted region skeleton;
Using l1Intermediate value skeleton technique, calculates candidate feature point region, obtains skeleton, as shown in Figure 7b.
3) resampling is carried out to candidate point using skeleton, obtains characteristic point;
According to step 2) skeletal point on the skeleton that obtains and skeleton, marked off in the middle of former and later two nearest skeletal points
One zonule, to the candidate feature point in the region that falls, selects one, and used as characteristic point, effect is as shown in Figure 7 c.
As shown in Figure 2 a, what is shown is method when resampling is carried out to edge point, and each skeleton point sampling obtains one
Characteristic point, to skeletal pointWhen being sampled, with its forward and backward skeletal pointCentered on, perpendicular to framework construction
Two plane lj-1、lj+1, to falling in two candidate feature points of plane institute clip space, selection one is projected toPlace plane is thrown
Shadow point most long, as the characteristic point for sampling;As shown in Figure 2 b, expression is method that angle steel joint carries out resampling, equally
Be also by the angle point in skeleton centered on, the candidate feature point in the region folded by the plane constituted in skeletal point adjacent thereto that will fall
Pick out, first calculate a little with the vector of skeleton angle point, addition of vectors obtains a principal direction, one point of reselection, its
The projection of corresponding vector to principal vector is most long, and this point is chosen as angle point.
4) many normal vectors are assigned to characteristic point;
According to step 3) judge the normal vector quantity that the point should possess, by near k Neighbor Points of the point, (i.e. k is near
Adjoint point) K-means clusters are carried out, the normal vector of the point in each class is averaged, obtain many normal vectors of the point.Effect is such as
Shown in Fig. 8.
5) a cloud normal vector is filtered based on joint bilateral filtering framework;
Need first to build guiding thresholding, this method will guide thresholding to be set to guiding normal vector a little here, calculate guidance method
The process of vector is as shown in Figure 3.Each point and its k Neighbor Points are set to a block, the point that each block is included is marked with yellow.When
When we will calculate the guiding normal vector of green point, the block comprising green point is all included.Its institute is calculated each block
Average normal vector a little, while calculating the average value of the vector difference of any two points in block, is defined as the continuous of each block
Property value (i.e. numeral in Fig. 3 below each figure), for weighing the vectorial uniformity of its internal point, continuity value is bigger, card
The bright block is across sharp features, and the point comprising another face is more.The invention aims to retain sharp features, distinguish
Go out the Different Plane of sharp features both sides, therefore it is the minimum block of continuity value that we need to select, i.e., do not span across as far as possible
The block of sharp features, by the average normal vector of this block, as the guiding normal vector of green point, i.e. the blue arrow institute of most right figure
Show.
Joint bilateral filtering framework is filtered to a cloud normal vector, and its formula is as follows:
Wherein,It is piJ-th filtered normal vector of point (assumes piPoint has multiple normal vectors), pk∈N(pi) show
pkPoint is belonging to piNeighborhood of a point point, nklIt is pkNormal vector before l-th filtering of point (assumes pkPoint has multiple normal vectors),It is alternate position spike weight, works as pkPoint and piWhen the Euclidean distance of point is more remote, weight is smaller;It is normal vector difference weight, works as pkThe guiding normal vector g of pointklWith piThe guidance method of point
Vectorial gijNormal vector difference it is bigger when, weight is smaller.WijIt is normalization factor, it is ensured that filtered normal vector mould a length of 1.
Work as pkPoint is when having multiple normal vectors, just select in its multiple normal vector with piThe point immediate normal vector of former normal vector
nkl, substitute into joint bilateral filtering formula and calculated, obtain piThe filtered normal vector of point.
6) position is updated based on filtered normal vector, obtains the point cloud model after denoising.
Based on filtered normal vectorI represents the number at model midpoint, has n point;
J represents the vectorial number that each point has, when i is general point, diWhen=1, i point are characterized, di> 1, is carried out more to a position
Newly, the point cloud model after denoising is obtained, it is necessary to the optimization problem being defined as follows:
εpointupd=εposition+εnormal (2)
Wherein:
In formula, piRepresent the point position, p ' after updatingiRepresent piPosition before renewal, NiRepresent p 'iK neighbour's point sets,
p′kRepresent p 'iK Neighbor Points, w (i, k)=wp(i,k)wn(i, k), whereinIt is used for
Position weighing factor of the neighborhood point to central point is weighed, when a distance is bigger, weight is smaller;It is used for
Normal vector weighing factor of the neighborhood point to central point is weighed, when the normal vector of the neighborhood point differs bigger with the normal vector of central point
When, weight is smaller;The iterative formula that solution formula (2) is obtained is:
Wherein t and t+1 is the t times and the t+1 times iteration, thus obtains new point positionBy iteration it is multiple after
Point cloud model after to denoising.
The present invention proves its feasibility by experiment, can be widely used in each class model.Fig. 4 shows the present invention to people
Modeling type carries out the result after denoising, and the several work best with effect in point cloud denoising field at present Comparative result.It is right
In manikin, initial conditions are no noises, and we are that what is added here is that Gauss makes an uproar plus noise by artificial disturbance
Sound, Noise Criterion is 2.0%, it can be seen that our method can be while denoising, to keep local sharp special well
Levy, and be not in depression or be some point situations about cannot process that suspend.Fig. 5 shows the present invention to actually sweeping
The model retouched carries out the result after denoising, and the several work best with effect in point cloud denoising field at present Comparative result.
For actual scanning model, noise is, because the many-sided condition of precision, the scanning circumstance etc. of scanning device is produced, more to represent
Situation under our practical application scenes.The same present invention preferably while denoising can retain point compared to other work
Sharp feature.In two examples of Fig. 4 and Fig. 5, different colours represent Different Plane, therefore the result of denoising should be same flat
Solid colour is kept inside face as far as possible, and adjacent Different Plane color should be distinguished as far as possible, to protrude sharp features.From
We are it is obvious that the present invention can preferably accomplish that while denoising, the sharp features of reserving model are not by mould in figure
Paste falls.
The present invention has more preferable robustness, can preferably process large scale noise.As shown in fig. 6, working as Gaussian noise
When being added to 4% (fourth line), other related processing methods cannot obtain very good effect, but the method for the present invention still table
Now stablize.
In sum, the point cloud denoising method based on joint bilateral filtering and sharp features skeletal extraction of the invention, can
With while denoising effective using joint bilateral filtering, by extracting sharp features skeleton analysis model structure, retain sharp
Feature, is conducive to optimizing the model that there is large scale noise, be China of China in the popularization of 3D printing technique, for
The optimization of model quality provides help, with actual promotional value, is worthy to be popularized.
The examples of implementation of the above are only the preferred embodiments of the invention, not limit implementation model of the invention with this
Enclose, therefore the change that all shapes according to the present invention, principle are made, all should cover within the scope of the present invention.
Claims (7)
1. the point cloud denoising method of joint bilateral filtering and sharp features skeletal extraction is based on, it is characterised in that including following step
Suddenly:
1) candidate feature point is obtained using normal vector difference method;
2) to candidate point extracted region skeleton;
3) resampling is carried out to candidate point using skeleton, obtains characteristic point;
4) many normal vectors are assigned to characteristic point;
5) a cloud normal vector is filtered based on joint bilateral filtering framework;
6) position is updated based on filtered normal vector, obtains the point cloud model after denoising.
2. according to claim 1 based on the point cloud denoising method for combining bilateral filtering and sharp features skeletal extraction, its
It is characterised by:In step 1) in, candidate feature point is obtained using normal vector difference method, it is to calculate any two in each vertex neighborhood
The normal vector difference of point is simultaneously sued for peace, and when this summing value is more than predetermined threshold value, this point is just labeled as candidate feature point.
3. according to claim 1 based on the point cloud denoising method for combining bilateral filtering and sharp features skeletal extraction, its
It is characterised by:In step 2) in, it is to use l to candidate point extracted region skeleton1Intermediate value skeleton technique, to candidate feature point area
Domain is calculated, and obtains skeleton.
4. according to claim 1 based on the point cloud denoising method for combining bilateral filtering and sharp features skeletal extraction, its
It is characterised by:In step 3) in, resampling is carried out to candidate point using skeleton, characteristic point is obtained, according to step 2) obtain
Skeletal point on skeleton and skeleton, marks off a zonule, to falling in the region in the middle of two neighboring skeletal point
Candidate feature point, selects one, as characteristic point.
5. according to claim 1 based on the point cloud denoising method for combining bilateral filtering and sharp features skeletal extraction, its
It is characterised by:In step 4) in, many normal vectors are assigned to characteristic point, be according to step 3) judge the normal vector that the point should possess
Quantity, will be that k Neighbor Points carry out K-means clusters near k Neighbor Points of the point, and the normal vector of the point in each class is asked
Average value, obtains many normal vectors of the point.
6. according to claim 1 based on the point cloud denoising method for combining bilateral filtering and sharp features skeletal extraction, its
It is characterised by:In step 5) in, a cloud normal vector is filtered, it is necessary to first build aiming field based on joint bilateral filtering framework
Value, will guide thresholding to be set to guiding normal vector a little, each point and its k Neighbor Points is set into a block, by the block comprising the point
All include and;Each block is calculated its average normal vector a little, while calculating the vector difference of any two points in block
Average value, is defined as the continuity value of each block, and for weighing the vectorial uniformity of its internal point, continuity value is got over
Greatly, it was demonstrated that the block is across sharp features, and the point comprising another face is more;In order to retain sharp features, sharp spy is distinguished
The Different Plane of both sides is levied, it is necessary to what is selected is the minimum block of continuity value, i.e., the block of sharp features is not spanned across as far as possible, by this
The average normal vector of individual block, as the guiding normal vector of the point;
Joint bilateral filtering formula is as follows:
Wherein,It is piJ-th filtered normal vector of point, pk∈N(pi) show pkPoint is belonging to piNeighborhood of a point point, nklIt is pk
Normal vector before l-th filtering of point,It is alternate position spike weight, works as pkPoint and piPoint
When Euclidean distance is more remote, weight is smaller;It is normal vector difference weight, works as pkThat puts draws
Inducing defecation by enema and suppository vector gklWith piThe guiding normal vector g of pointijNormal vector difference it is bigger when, weight is smaller;WijIt is normalization factor, it is ensured that
Filtered normal vector mould a length of 1;
Work as pkPoint is when having multiple normal vectors, just select in its multiple normal vector with piThe immediate normal vector n of the former normal vector of pointkl, generation
Enter joint bilateral filtering formula to be calculated, obtain piThe filtered normal vector of point.
7. according to claim 1 based on the point cloud denoising method for combining bilateral filtering and sharp features skeletal extraction, its
It is characterised by:In step 6) in, based on filtered normal vectorI represents the individual of model midpoint
Number, has n point;J represents the vectorial number that each point has, when i is general point, diWhen=1, i point are characterized, di> 1,
A position is updated, the point cloud model after denoising is obtained, it is necessary to the optimization problem being defined as follows:
εpointupd=εposition+εnormal (2)
Wherein:
In formula, piRepresent the point position, p ' after updatingiRepresent piPosition before renewal, NiRepresent p 'iK neighbour's point sets, p'k
Represent p 'iK Neighbor Points, w (i, k)=wp(i,k)wn(i, k), whereinFor weighing
, to the position weighing factor of central point, when a distance is bigger, weight is smaller for amount neighborhood point;For weighing
Normal vector weighing factor of the neighborhood point to central point is measured, when the normal vector of the neighborhood point differs bigger with the normal vector of central point
When, weight is smaller;The iterative formula that solution formula (2) is obtained is:
Wherein t and t+1 is the t times and the t+1 times iteration, thus obtains new point positionBy iteration it is multiple after gone
Point cloud model after making an uproar.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611180960.6A CN106709883B (en) | 2016-12-20 | 2016-12-20 | Point cloud denoising method based on joint bilateral filtering and sharp features skeletal extraction |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201611180960.6A CN106709883B (en) | 2016-12-20 | 2016-12-20 | Point cloud denoising method based on joint bilateral filtering and sharp features skeletal extraction |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106709883A true CN106709883A (en) | 2017-05-24 |
CN106709883B CN106709883B (en) | 2019-10-18 |
Family
ID=58938568
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201611180960.6A Expired - Fee Related CN106709883B (en) | 2016-12-20 | 2016-12-20 | Point cloud denoising method based on joint bilateral filtering and sharp features skeletal extraction |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106709883B (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107862749A (en) * | 2017-11-27 | 2018-03-30 | 华南理工大学 | One kind point cloud geometric detail feature minimizing technology |
CN109767463A (en) * | 2019-01-09 | 2019-05-17 | 重庆理工大学 | A kind of three-dimensional point cloud autoegistration method |
CN110232698A (en) * | 2019-04-20 | 2019-09-13 | 北京工业大学 | One kind is based on model segmentation and L1The successive frame 3 D human body curve skeleton extracting method of intermediate value skeleton |
CN112396567A (en) * | 2020-11-26 | 2021-02-23 | 河北工业大学 | Scattered point cloud denoising method based on two-step method of normal correction and position filtering |
CN112419164A (en) * | 2019-08-23 | 2021-02-26 | 南京理工大学 | Curvature and neighborhood reconstruction-based weighted guide point cloud model denoising method |
WO2021129317A1 (en) * | 2019-12-26 | 2021-07-01 | 华南理工大学 | Point cloud smoothing filtering method based on normal vector |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120195492A1 (en) * | 2011-01-28 | 2012-08-02 | Sony Corporation | Method and apparatus for generating a dense depth map using an adaptive joint bilateral filter |
CN102938824A (en) * | 2012-06-05 | 2013-02-20 | 李英明 | Video denoising method based on improved self-adaptive bilateral filter |
US20160163101A1 (en) * | 2014-12-04 | 2016-06-09 | Korea Advanced Institute Of Science And Technology | Apparatus and method for reconstructing skeletal image |
CN105913396A (en) * | 2016-04-11 | 2016-08-31 | 湖南源信光电科技有限公司 | Noise estimation-based image edge preservation mixed de-noising method |
US20160300328A1 (en) * | 2014-01-17 | 2016-10-13 | Tencent Technology (Shenzhen) Co., Ltd. | Method and apparatus for implementing image denoising |
KR101666449B1 (en) * | 2016-05-13 | 2016-10-14 | 한국건설기술연구원 | Apparatus for reverse engineering sweeping shape using feature template similarity based on 3 dimentional point cloud and method thereof |
-
2016
- 2016-12-20 CN CN201611180960.6A patent/CN106709883B/en not_active Expired - Fee Related
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20120195492A1 (en) * | 2011-01-28 | 2012-08-02 | Sony Corporation | Method and apparatus for generating a dense depth map using an adaptive joint bilateral filter |
CN102938824A (en) * | 2012-06-05 | 2013-02-20 | 李英明 | Video denoising method based on improved self-adaptive bilateral filter |
US20160300328A1 (en) * | 2014-01-17 | 2016-10-13 | Tencent Technology (Shenzhen) Co., Ltd. | Method and apparatus for implementing image denoising |
US20160163101A1 (en) * | 2014-12-04 | 2016-06-09 | Korea Advanced Institute Of Science And Technology | Apparatus and method for reconstructing skeletal image |
CN105913396A (en) * | 2016-04-11 | 2016-08-31 | 湖南源信光电科技有限公司 | Noise estimation-based image edge preservation mixed de-noising method |
KR101666449B1 (en) * | 2016-05-13 | 2016-10-14 | 한국건설기술연구원 | Apparatus for reverse engineering sweeping shape using feature template similarity based on 3 dimentional point cloud and method thereof |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107862749A (en) * | 2017-11-27 | 2018-03-30 | 华南理工大学 | One kind point cloud geometric detail feature minimizing technology |
CN107862749B (en) * | 2017-11-27 | 2020-02-18 | 华南理工大学 | Point cloud geometric detail feature removing method |
CN109767463A (en) * | 2019-01-09 | 2019-05-17 | 重庆理工大学 | A kind of three-dimensional point cloud autoegistration method |
CN109767463B (en) * | 2019-01-09 | 2021-04-13 | 重庆理工大学 | Automatic registration method for three-dimensional point cloud |
CN110232698A (en) * | 2019-04-20 | 2019-09-13 | 北京工业大学 | One kind is based on model segmentation and L1The successive frame 3 D human body curve skeleton extracting method of intermediate value skeleton |
CN110232698B (en) * | 2019-04-20 | 2021-09-17 | 北京工业大学 | Based on model segmentation and L1Method for extracting continuous frame three-dimensional human body curve skeleton of median skeleton |
CN112419164A (en) * | 2019-08-23 | 2021-02-26 | 南京理工大学 | Curvature and neighborhood reconstruction-based weighted guide point cloud model denoising method |
WO2021129317A1 (en) * | 2019-12-26 | 2021-07-01 | 华南理工大学 | Point cloud smoothing filtering method based on normal vector |
CN112396567A (en) * | 2020-11-26 | 2021-02-23 | 河北工业大学 | Scattered point cloud denoising method based on two-step method of normal correction and position filtering |
CN112396567B (en) * | 2020-11-26 | 2023-06-06 | 河北工业大学 | Scattered point cloud denoising method based on normal correction and position filtering two-step method |
Also Published As
Publication number | Publication date |
---|---|
CN106709883B (en) | 2019-10-18 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106709883A (en) | Point cloud denoising method based on joint bilateral filtering and sharp feature skeleton extraction | |
CN109872397B (en) | Three-dimensional reconstruction method of airplane parts based on multi-view stereo vision | |
CN107093205B (en) | A kind of three-dimensional space building window detection method for reconstructing based on unmanned plane image | |
CN105551028B (en) | A kind of method and system of the geographical spatial data dynamic renewal based on remote sensing image | |
CN109241871A (en) | A kind of public domain stream of people's tracking based on video data | |
CN107025687B (en) | A kind of view-based access control model relies on the historical relic line chart method for drafting of Curvature Estimate | |
CN106504277B (en) | A kind of improved ICP point cloud autoegistration method | |
CN105574933B (en) | The comprehensive profile accurate Drawing method of object | |
JP2008530595A (en) | Method and apparatus for improving digital elevation model (DEM) for terrain modeling | |
CN106056606A (en) | Image processing method and device | |
JP2008530594A (en) | Method and apparatus for distinguishing buildings from vegetation for terrain modeling | |
CN107657659A (en) | The Manhattan construction method for automatic modeling of scanning three-dimensional point cloud is fitted based on cuboid | |
CN108460833A (en) | A kind of information platform building traditional architecture digital protection and reparation based on BIM | |
CN110379004A (en) | The method that a kind of pair of oblique photograph achievement carries out terrain classification and singulation is extracted | |
CN107978017A (en) | Doors structure fast modeling method based on wire extraction | |
CN104166992B (en) | Perception of content binocular image Zoom method based on distortion of the mesh | |
CN104484679B (en) | Non- standard rifle shooting warhead mark image automatic identifying method | |
CN105354812A (en) | Method for identifying profile interaction based on multi-Kinect collaboration depth threshold segmentation algorithm | |
CN109558801B (en) | Road network extraction method, medium, computer equipment and system | |
JP6146731B2 (en) | Coordinate correction apparatus, coordinate correction program, and coordinate correction method | |
CN108062758B (en) | A kind of crowd's generation emulation mode and system based on image segmentation algorithm | |
CN110517348A (en) | Target object three-dimensional point cloud method for reconstructing based on display foreground segmentation | |
CN105631849B (en) | The change detecting method and device of target polygon | |
CN107862749A (en) | One kind point cloud geometric detail feature minimizing technology | |
JP5295044B2 (en) | Method and program for extracting mask image and method and program for constructing voxel data |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191018 |