CN105096268B - One kind point cloud denoising smooth method - Google Patents
One kind point cloud denoising smooth method Download PDFInfo
- Publication number
- CN105096268B CN105096268B CN201510408335.1A CN201510408335A CN105096268B CN 105096268 B CN105096268 B CN 105096268B CN 201510408335 A CN201510408335 A CN 201510408335A CN 105096268 B CN105096268 B CN 105096268B
- Authority
- CN
- China
- Prior art keywords
- point
- formula
- point cloud
- minpts
- cloud model
- 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.)
- Expired - Fee Related
Links
Landscapes
- Image Processing (AREA)
Abstract
The invention discloses one kind to put cloud denoising smooth method, denoising is carried out to a cloud using density self-adapting clustering method, the point cloud after denoising is carried out using bilateral filtering algorithm smooth, and the point cloud model to obtaining after smooth is rebuild, threedimensional model after being rebuild, the initial radium in density self-adapting clustering method and minimum Neighborhood Number purpose value are determined by calculating, the defects of avoiding original clustering method continuous according to cluster result adjusting parameter, so as to improve processing speed.
Description
Technical field
The invention belongs to computer graphics techniques field, is related to a kind of point cloud denoising smooth method.
Background technology
With the sharp increase of model polygon complexity, the advantage of point cloud model is more and more obvious, using a cloud as research object
Graphics increasingly attract attention.With the rapid development of DATA REASONING technology, the measurement accuracy of spatial digitizer constantly carries
Height, the point collected milk up collection, and contain abundant information in kind.But due to the adaptation of scanner itself
The influence of property, human factor, environmental factor and measuring method etc., the data of collection always include noise, it is impossible to accurately show quilt
Survey the surface information of crops.The presence of noise spot can influence the precision of subsequent characteristics point extraction and rebuild the matter of threedimensional model
Amount, cause reconstruct curve, curved surface rough, reduce the precision of reconstruction model.
Traditional point cloud denoising method is more using the denoising method of triangle grid model and cloud data, with Laplce
When operator is adjusted to triangular plate method arrow, because this method is isotropic, key character blooming be present；Respectively to different
Although property Mesh Smoothing Algorithm can keep geometric properties, generally use high-order geometry flow, algorithm complex is higher, some situations
Under can also cause the deformation and distortion of grid model；In step-by-step processing, the fuzzy C-mean algorithm method for large scale noise remove
To noise-sensitive, noise quality may be reduced.Although the above method can reach certain effect, total existing characteristics mould
Paste, the deformation of model and distortion and the problem of to noise-sensitive, influence the precision of subsequent characteristics extraction and the quality of reconstruction model.
The content of the invention
For above-mentioned problems of the prior art and defect, it is an object of the present invention to provide one kind to put cloud denoising
Smoothing method.
To achieve these goals, the present invention adopts the following technical scheme that：
One kind point cloud denoising smooth method, specifically includes following steps：
Step 1：Density self-adapting cluster analysis is carried out to three-dimensional point cloud model, obtains the point cloud model after denoising, it has
Body implementation method is as follows：
Step 1.1：Three-dimensional point cloud model is imported into the three-dimensional system of coordinate using x, y, z as reference axis, calculates auto-adaptive parameter
Initial radium e and minimum neighborhood number MinPts, and assign initial value 1 for class-mark ClusterID；
Step 1.2：A set D will be used as a little in three-dimensional point cloud model, reading one from D has not visited
Object P, and search all objects reachable from P density in D on e and MinPts；
Step 1.3：If P is kernel object, i.e., the sample points in point P e neighborhoods are more than or equal to MinPts, then object P
Class-mark be entered as ClusterID；If P is not kernel object, as border object, then object P class-mark is entered as 0, table
It is isolated noise spot to show object P；
Step 1.4：Continued search for since the reachable object of the object P density that step 1.2 obtains, until being without object
Only, it is ClusterID by the reachable object value of all density, ClusterID adds 1 certainly, and goes to step 1.2, until to be sorted
All objects are all accessed in set D；
Step 2：What step 1 was obtained removes the point cloud progress bilateral filtering after noise, the point cloud mould after obtaining smoothly
Type；
Step 3：The point cloud model obtained for step 2, three after being rebuild using the generation of Delaunay Triangulation method
Dimension module.
Specifically, in the step 1.1：
The computational methods of the auto-adaptive parameter initial radium e are as follows：
Any two points p in three-dimensional point cloud model is calculated according to formula 1iAnd pjBetween Euclidean distance dist (i, j)：
Wherein, (xi,yi,zi) and (xj,yj,zj) point p is represented respectivelyiAnd pjCoordinate；
Dist (i, j) maximum maxdist and minimum value mindist is calculated according to formula 2 and formula 3, according to formula 4
Try to achieve distance interval distrange；
Maxdist=Max dist (i, j) | 0≤i<n,0≤j<N } formula 2
Mindist=Min dist (i, j) | 0≤i<n,0≤j<N } formula 3
Distrange=maxdist-mindist formulas 4
Wherein, n represents the number at three-dimensional point cloud model midpoint；
Ten sections will be equidistantly divided into apart from interval distrange, the endpoint value being respectively segmented uses d respectively from small to large0,d1,...,
d10Represent, number ps of the statistics dist (i, j) in every segment limitk, 0≤k<10, wherein, subscript k represents the above-mentioned section being respectively segmented
Number, note number pkIn maximum be pm, 0≤m<10, then auto-adaptive parameter initial radium e is tried to achieve according to formula 5：
The computational methods of the minimum neighborhood number MinPts are as follows：
After auto-adaptive parameter initial radium e is determined, the initial value for setting MinPts is 1；Point cloud mould is counted according to formula 6
Arbitrfary point p in typeiNeighborhood count out pNumi, 0≤i<N counts all neighborhoods of a point in point cloud model according to formula 7 and counted
The number of point of the mesh more than MinPts, as pNum；
pNumi=count { dist (i, j)<e|0≤j<N } formula 6
PNum=count { pNumi≥MinPts|0≤i<N } formula 7
Minimum neighborhood number MinPts is incrementally increased, every time incrementally 1, with minimum neighborhood number MinPts increase,
PNum can be gradually decreased and tended towards stability, and calculate pNum every time afterwards compared with the preceding result once calculated, selection pNum is not
Minimum neighborhood number when changing again is as MinPts.
Specifically, the concrete methods of realizing of the step 2 is as follows：
Step 2.1：Each data point q of point cloud after the removal noise obtained for step 1i, obtain its m
Neighbor point kij, j=1,2 ..., m, m Neighbor Points be calculated as follows：
Data point q is calculated according to formula 1iWith other in a cloud distance a little, then by above-mentioned distance press from it is small to
Big sequence, m point is as m Neighbor Points before interception.
Step 2.2：The parameter x=of smoothing filter function is obtained for each neighbor point | | qi-kij| | and feature keeps power
The parameter of weight functionWherein x is point qiTo neighbor point kijDistance, y is point qiWith the distance vector of neighbor point
qi-kijWith the normal directionInner product；
Step 2.3：Smoothing filter function W is calculated according to formula 8 and formula 9c(x) and feature keeps weighting function Ws
(y)；
Wherein：σc- spatial domain weight；σs- property field weight；
Step 2.4：By WcAnd W (x)s(y) formula 10 is substituted into, calculates bilateral filtering factor-alpha；
Wherein：N(qi) it is data point qiNeighborhood.
Step 2.5：New data point after being calculated after filtering according to formula 11；
Wherein：qnewFor filtered data point；qoldData point after the denoising obtained for step 1；α adds for bilateral filtering
Weight factor；For data point qoldPer unit system arrow.
Compared with prior art, the present invention has following technique effect：
1st, the present invention carries out denoising using density self-adapting clustering method to a cloud, using bilateral filtering algorithm to going
Model after making an uproar carries out smooth, and the point cloud model to obtaining after smooth is rebuild, the threedimensional model after being rebuild, fully
Using density clustering algorithm to insensitive for noise, the characteristics of arbitrary shape, big group cluster can be found and bilateral filtering holding point cloud
The characteristics of feature, realize and denoising smooth is carried out in initial data while retains the more features of cloud data, pass through contrast
Experiment finds that the above method has preferable denoising smooth effect.
2nd, the initial radium in density self-adapting clustering method and minimum Neighborhood Number purpose value are determined by calculating, kept away
The defects of having exempted from original clustering method continuous according to cluster result adjusting parameter, so as to improve processing speed；Cause three simultaneously
The denoising smooth effect of dimension point cloud is more preferable.
3rd, when progress bilateral filtering is smooth, it is contemplated that influence of the neighborhood point to current point, avoided smooth phenomenon
Occur.
Brief description of the drawings
Fig. 1 is the three-dimensional point cloud model design sketch of regular object；Fig. 1 (a) is original point cloud model；Fig. 1 (b) is using this
The denoising glossy effect figure that the method for invention obtains；Fig. 1 (c) is that original point cloud rebuilds effect；Fig. 1 (d) is to use present invention side
The reconstruction design sketch that method obtains.
Fig. 2 is the three-dimensional point cloud model design sketch of irregularly shaped object；Fig. 2 (a) is original point cloud model；Fig. 2 (b) is contrast
Test obtained design sketch；Fig. 2 (c) is the denoising glossy effect that the inventive method obtains；Fig. 2 (d) is that original point cloud rebuilds effect
Fruit is schemed；Fig. 2 (e) is the reconstruction design sketch that contrast experiment obtains；Fig. 2 (f) is the reconstruction effect obtained using the inventive method.
Fig. 3 is flow chart of the method for the present invention.
With reference to the accompanying drawings and detailed description to the solution of the present invention explanation and illustration in further detail.
Embodiment
Above-mentioned technical proposal is deferred to, referring to Fig. 3, point cloud denoising smooth method of the invention, specifically includes following steps：
Step 1：Density self-adapting cluster analysis is carried out to three-dimensional point cloud model, obtains the point cloud model after denoising.It has
Body implementation method is as follows：
Step 1.1：The three-dimensional point cloud model for coming the acquisition of card scanstation2 three-dimensional laser scanners is imported with x, y, z
For the three-dimensional system of coordinate of reference axis, auto-adaptive parameter initial radium e and minimum neighborhood number MinPts is calculated, and be class-mark
ClusterID assigns initial value 1.Wherein, auto-adaptive parameter initial radium e computational methods are as follows：
Any two points p in three-dimensional point cloud model is calculated according to formula (1)iAnd pjBetween Euclidean distance dist (i, j)：
Wherein, (xi,yi,zi) and (xj,yj,zj) point p is represented respectivelyiAnd pjCoordinate.
Dist (i, j) maximum maxdist and minimum value mindist is calculated according to formula (2) and (3), according to formula
(4) distance interval distrange is tried to achieve.
Maxdist=Max dist (i, j) | 0≤i<n,0≤j<n} (2)
Mindist=Min dist (i, j) | 0≤i<n,0≤j<n} (3)
Distrange=maxdist-mindist (4)
Wherein, n represents the number at three-dimensional point cloud model midpoint.
Ten sections will be equidistantly divided into apart from interval distrange, the endpoint value being respectively segmented uses d respectively from small to large0,d1,...,
d10Represent, the number p that statistics dist (i, j) occurs in every segment limitk(0≤k<10), wherein, subscript k represents above-mentioned each point
The segment number of section, note number pkIn maximum be pm(0≤m<10), then auto-adaptive parameter initial radium e is tried to achieve according to formula (5)：
Minimum neighborhood number MinPts computational methods are as follows：
After auto-adaptive parameter initial radium e is determined, the initial value for setting MinPts is 1；Point cloud is counted according to formula (6)
Arbitrfary point p in modeliNeighborhood count out pNumi, (0≤i<N) according to formula (7) count in point cloud model neighbour a little
Domain is counted out the number of the point more than MinPts, as pNum；
pNumi=count { dist (i, j)<e|0≤j<n} (6)
PNum=count { pNumi≥MinPts|0≤i<n} (7)
Minimum neighborhood number MinPts (every time be incremented by 1) is incrementally increased, with minimum neighborhood number MinPts increase,
PNum can be gradually decreased and tended towards stability, and calculate pNum every time afterwards compared with the preceding result once calculated, selection pNum is not
Minimum neighborhood number when changing again is as MinPts.
In traditional algorithm, it is necessary to according to Clustering Effect after initial radium e and minimum neighborhood number MinPts initializations
Constantly adjustment is to obtain optimal Clustering Effect, and parameter adjustment is comparatively laborious time-consuming, so carrying out the two by the above method
The adaptive adjustment of parameter, the continuous adjustment of parameter is avoided while ideal effect is obtained.
Step 1.2：A set D will be used as a little in three-dimensional point cloud model, reading one from D has not visited
Object P, and search all objects reachable from P density in D on e and MinPts.
Step 1.3：If P is kernel object, i.e., the sample points in point P e neighborhoods are more than or equal to MinPts, then object P
Class-mark be entered as ClusterID；If P is not kernel object, as border object, then object P class-mark is entered as 0, table
It is isolated noise spot to show object P.
Step 1.4：Continued search for since the reachable object of the object P density that step 1.2 obtains, until being without object
Only, it is ClusterID by the reachable object value of all density, ClusterID adds 1 certainly, and goes to step 1.2, until to be sorted
All objects are all accessed in set D.
Step 2：What step 1 was obtained removes the point cloud progress bilateral filtering after noise, the point cloud mould after obtaining smoothly
Type.Concrete methods of realizing is as follows：
Step 2.1：Each data point q of point cloud after the removal noise obtained for step 1i, obtain its m
Neighbor point kij, j=1,2 ..., m, m Neighbor Points be calculated as follows：
Data point q is calculated according to formula (1)iWith other in a cloud distance a little, then above-mentioned distance is pressed from small
To big sequence, m point is as data point q before interceptioniM Neighbor Points.
Step 2.2：The parameter x=of smoothing filter function is obtained for each neighbor point | | qi-kij| | and feature keeps power
The parameter of weight functionWherein x is point qiTo neighbor point kijDistance, y is point qiWith the distance of neighbor point to
Measure qi-kijWith the normal directionInner product；
Step 2.3：Smoothing filter function W is calculated according to formula (8) and formula (9)c(x) and feature keeps weighting function
Ws(y)；
Wherein：σc- spatial domain weight, it is data point qiDistance to neighbor point is bigger to the factor of influence of the point, its value
Effect is better, but feature holding capacity is weaker, is for used three-dimensional point cloud model, its span in the present invention
0.001-0.01；σs- property field weight, it is data point qiTo neighbor point distance vector in the normal directionOn projection logarithm
Strong point qiFactor of influence, its value it is more big can protection feature information, but reduce smoothing capability simultaneously, pin in the present invention
To used three-dimensional point cloud model, its span is 0.001-0.01.
Step 2.4：By WcAnd W (x)s(y) formula (10) is substituted into, calculates bilateral filtering factor-alpha；
Wherein：N(qi) it is data point qiNeighborhood.
Step 2.5：New data point after being calculated after filtering according to formula (11).
Wherein：qnewFor filtered data point；qoldData point after the denoising obtained for step 1；α adds for bilateral filtering
Weight factor；For data point qoldPer unit system arrow.
Step 3：The point cloud model obtained for step 2, three after being rebuild using the generation of Delaunay Triangulation method
Dimension module.
Experimental result
Experiment 1
Emulation experiment is carried out to the three-dimensional point cloud model of regular object, it is original shown in its original point cloud model such as Fig. 1 (a)
Point cloud is rebuild shown in design sketch such as Fig. 1 (c)；Denoising smooth is carried out to three-dimensional point cloud model using the method for the present invention, gone
Make an uproar it is smooth after design sketch, as described in Fig. 1 (b), shown in reconstruction design sketch such as Fig. 1 (d) after denoising smooth, by Fig. 1 (a) and
Fig. 1 (b), and Fig. 1 (c) and Fig. 1 (d) contrast are found, three-dimensional point cloud mould can be preferably realized using the method for the present invention
The purpose of type denoising smooth.
Experiment 2
The method of three-dimensional point cloud model and the present invention to irregularly shaped object carries out emulation experiment and obtains simulation result, contrasts
Experiment uses the bilateral filtering method in adaptive clustering scheme and document [2] in document [1] to carry out denoising smooth, is imitated
True result.
Shown in original point cloud model such as Fig. 2 (a), original point cloud is rebuild shown in design sketch such as Fig. 2 (d)；
Denoising smooth is carried out to three-dimensional point cloud model using the method in contrast experiment, obtains the effect after denoising smooth
Figure, as shown in Fig. 2 (b), shown in reconstruction design sketch such as Fig. 2 (e) after denoising smooth,
Denoising smooth is carried out to three-dimensional point cloud model using the method for the present invention, obtains the design sketch after denoising smooth, such as
Shown in Fig. 2 (c), shown in reconstruction design sketch such as Fig. 2 (f) after denoising smooth,
Original point cloud it can be seen from partial enlargement effect (oval part) in Fig. 2 (d), Fig. 2 (e) and Fig. 2 (f)
Rebuilding chart face has part raised；Still there is part projection on reconstruction surface in contrast experiment, but rebuilds effect than original point
Yun Hao；And the reconstruction surface being obtained by the present invention is adopted compared with contrast experiment, the smooth no projection in surface.
Bibliography：
[1] application [J] of Zhang Qiaoying, Chen Hao, Zhu Shuan density clustering algorithms in continuously distributed cloud denoising is geographical empty
Between information, 2011,9 (6)
[2] Du little Yan, Jiang Xiaofeng, Hao Chuangang, wait bilateral filtering Denoising Algorithm [J] the computer applications of point cloud models with
Software, 2010,27 (7) .245-246.
Claims (2)
1. one kind point cloud denoising smooth method, it is characterised in that specifically include following steps：
Step 1：Density self-adapting cluster analysis is carried out to three-dimensional point cloud model, obtains the point cloud model after denoising, it is specific real
Existing method is as follows：
Step 1.1：Three-dimensional point cloud model is imported into the three-dimensional system of coordinate using x, y, z as reference axis, it is initial to calculate auto-adaptive parameter
Radius e and minimum neighborhood number MinPts, and assign initial value 1 for class-mark ClusterID；
Step 1.2：A set D will be used as a little in three-dimensional point cloud model, an object having not visited is read from D
P, and search all objects reachable from P density in D on e and MinPts；
Step 1.3：If P is kernel object, i.e., the sample points in point P e neighborhoods are more than or equal to MinPts, then object P class
Number it is entered as ClusterID；If P is not kernel object, as border object, then object P class-mark is entered as 0, expression pair
As P is isolated noise spot；
Step 1.4：Continued search for since the reachable object of the object P density that step 1.2 obtains, will untill without object
The reachable object value of all density is ClusterID, and ClusterID adds 1 certainly, and goes to step 1.2, until set D to be sorted
In all objects be all accessed；
Step 2：What step 1 was obtained removes the point cloud progress bilateral filtering after noise, the point cloud model after obtaining smoothly；
Step 3：The point cloud model obtained for step 2, the three-dimensional mould after being rebuild using the generation of Delaunay Triangulation method
Type；
In the step 1.1：
The computational methods of the auto-adaptive parameter initial radium e are as follows：
Any two points p in three-dimensional point cloud model is calculated according to formula 1iAnd pjBetween Euclidean distance dist (i, j)：
Wherein, (xi,yi,zi) and (xj,yj,zj) point p is represented respectivelyiAnd pjCoordinate；
Dist (i, j) maximum maxdist and minimum value mindist is calculated according to formula 2 and formula 3, is tried to achieve according to formula 4
Distance interval distrange；
Maxdist=Max dist (i, j) | and 0≤i ＜ n, 0≤j ＜ n } formula 2
Mindist=Min dist (i, j) | and 0≤i ＜ n, 0≤j ＜ n } formula 3
Distrange=maxdist-mindist formulas 4
Wherein, n represents the number at three-dimensional point cloud model midpoint；
Ten sections will be equidistantly divided into apart from interval distrange, the endpoint value being respectively segmented uses d respectively from small to large0,d1,...,d10Table
Show, number ps of the statistics dist (i, j) in every segment limitk, 0≤k ＜ 10, wherein, subscript k represents the above-mentioned segment number being respectively segmented,
Remember number pkIn maximum be pm, 0≤m ＜ 10, then auto-adaptive parameter initial radium e is tried to achieve according to formula 5：
The computational methods of the minimum neighborhood number MinPts are as follows：
After auto-adaptive parameter initial radium e is determined, the initial value for setting MinPts is 1；Counted according to formula 6 in point cloud model
Arbitrfary point piNeighborhood count out pNumi, 0≤i<N counts all neighborhoods of a point in point cloud model according to formula 7 and counted out greatly
In the number of MinPts point, as pNum；
pNumi=count dist (i, j) ＜ e | and 0≤j ＜ n } formula 6
PNum=count { pNumi>=MinPts | 0≤i ＜ n } formula 7
Minimum neighborhood number MinPts is incrementally increased, every time incrementally 1, with minimum neighborhood number MinPts increase, pNum meetings
Gradually decrease and tend towards stability, calculate pNum every time afterwards compared with the preceding result once calculated, selection pNum no longer changes
When minimum neighborhood number as MinPts.
2. point cloud denoising smooth method as claimed in claim 1, it is characterised in that the concrete methods of realizing of the step 2 is such as
Under：
Step 2.1：Each data point q of point cloud after the removal noise obtained for step 1i, it is individual neighbouring to obtain its m
Point kij, j=1,2 ..., m, m Neighbor Points be calculated as follows：
Data point q is calculated according to formula 1iWith other in a cloud distance a little, then by above-mentioned distance by arranging from small to large
Sequence, m point is as m Neighbor Points before interception；
Step 2.2：The parameter x=of smoothing filter function is obtained for each neighbor point | | qi-kij| | and feature keeps weight letter
Several parametersWherein x is point qiTo neighbor point kijDistance, y is point qiWith the distance vector q of neighbor pointi-
kijWith the normal directionInner product；
Step 2.3：Smoothing filter function W is calculated according to formula 8 and formula 9c(x) and feature keeps weighting function Ws(y)；
Wherein：σc- spatial domain weight；σs- property field weight；
Step 2.4：By WcAnd W (x)s(y) formula 10 is substituted into, calculates bilateral filtering factor-alpha；
Wherein：N(qi) it is data point qiNeighborhood；
Step 2.5：New data point after being calculated after filtering according to formula 11；
Wherein：qnewFor filtered data point；qoldData point after the denoising obtained for step 1；α be bilateral filtering weight because
Son；For data point qoldPer unit system arrow.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510408335.1A CN105096268B (en) | 2015-07-13 | 2015-07-13 | One kind point cloud denoising smooth method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510408335.1A CN105096268B (en) | 2015-07-13 | 2015-07-13 | One kind point cloud denoising smooth method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105096268A CN105096268A (en) | 2015-11-25 |
| CN105096268B true CN105096268B (en) | 2018-02-02 |
Family
ID=54576619
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510408335.1A Expired - Fee Related CN105096268B (en) | 2015-07-13 | 2015-07-13 | One kind point cloud denoising smooth method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105096268B (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106874925A (en) * | 2015-12-14 | 2017-06-20 | 阿里巴巴集团控股有限公司 | object grouping method, model training method and device |
| CN107341768B (en) * | 2016-04-29 | 2022-03-11 | 微软技术许可有限责任公司 | Grid noise reduction |
| CN106447624A (en) * | 2016-08-31 | 2017-02-22 | 上海交通大学 | L0 norm-based three-dimensional grid denoising method |
| CN106767710A (en) * | 2016-12-22 | 2017-05-31 | 上海华测导航技术股份有限公司 | A kind of Earth Volume of Road Engineering measuring method and system |
| CN106846272A (en) * | 2017-01-18 | 2017-06-13 | 西安工程大学 | A kind of denoising compressing method of point cloud model |
| CN106918813B (en) * | 2017-03-08 | 2019-04-30 | 浙江大学 | A kind of three-dimensional sonar point cloud chart image intensifying method based on distance statistics |
| CN107123163A (en) * | 2017-04-25 | 2017-09-01 | 无锡中科智能农业发展有限责任公司 | A kind of plant three-dimensional reconstruction system based on various visual angles stereoscopic vision |
| CN108648156B (en) * | 2018-05-08 | 2021-04-16 | 北京邮电大学 | Method and device for marking stray points in point cloud data, electronic equipment and storage medium |
| CN109035224B (en) * | 2018-07-11 | 2021-11-09 | 哈尔滨工程大学 | Submarine pipeline detection and three-dimensional reconstruction method based on multi-beam point cloud |
| CN109146817B (en) * | 2018-08-23 | 2021-07-09 | 西安工业大学 | Noise processing method for non-iterative single object scattered point cloud data |
| CN109767391A (en) * | 2018-12-03 | 2019-05-17 | 深圳市华讯方舟太赫兹科技有限公司 | Point cloud denoising method, image processing equipment and the device with store function |
| CN109712174B (en) * | 2018-12-25 | 2020-12-15 | 湖南大学 | Point cloud misregistration filtering method and system for three-dimensional measurement of complex special-shaped curved surface robot |
| CN109934120B (en) * | 2019-02-20 | 2021-04-23 | 东华理工大学 | Step-by-step point cloud noise removing method based on space density and clustering |
| CN110349094A (en) * | 2019-06-12 | 2019-10-18 | 西安工程大学 | It is peeled off based on statistics and the 3D point cloud denoising method of adaptive bilateral mixed filtering |
| CN111105381B (en) * | 2020-02-07 | 2023-03-31 | 武汉玄景科技有限公司 | Dense point cloud smoothing method based on spherical model |
| CN112102178A (en) * | 2020-07-29 | 2020-12-18 | 深圳市菲森科技有限公司 | Point cloud feature-preserving denoising method and device, electronic equipment and storage medium |
| CN116310360B (en) * | 2023-05-18 | 2023-08-18 | 实德电气集团有限公司 | Reactor surface defect detection method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103793937A (en) * | 2013-12-10 | 2014-05-14 | 中山大学深圳研究院 | Density clustering-based three-dimensional geometrical model simplification method and device |
| CN104239556A (en) * | 2014-09-25 | 2014-12-24 | 西安理工大学 | Density clustering-based self-adaptive trajectory prediction method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9767598B2 (en) * | 2012-05-31 | 2017-09-19 | Microsoft Technology Licensing, Llc | Smoothing and robust normal estimation for 3D point clouds |
-
2015
- 2015-07-13 CN CN201510408335.1A patent/CN105096268B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103793937A (en) * | 2013-12-10 | 2014-05-14 | 中山大学深圳研究院 | Density clustering-based three-dimensional geometrical model simplification method and device |
| CN104239556A (en) * | 2014-09-25 | 2014-12-24 | 西安理工大学 | Density clustering-based self-adaptive trajectory prediction method |
Non-Patent Citations (3)
| Title |
|---|
| SA．DBSCAN：一种自适应基于密度聚类算法;夏鲁宁 等;《中国科学院研究生院学报》;20090715;第26卷(第4期);530-538 * |
| 密度聚类算法在连续分布点云去噪中的应用;张巧英 等;《地理空间信息》;20111228;第9卷(第6期);摘要、第3节、图5 * |
| 鲁棒的模糊 C 均值和点云双边滤波去噪;王丽辉 等;《北京交通大学学报》;20080415;第32卷(第2期);第1节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105096268A (en) | 2015-11-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105096268B (en) | One kind point cloud denoising smooth method | |
| CN103853840B (en) | Filter method of nonuniform unorganized-point cloud data | |
| CN108303037B (en) | Method and device for detecting workpiece surface shape difference based on point cloud analysis | |
| CN110807781B (en) | Point cloud simplifying method for retaining details and boundary characteristics | |
| CN115222625A (en) | Laser radar point cloud denoising method based on multi-scale noise | |
| CN108320293A (en) | A kind of combination improves the quick point cloud boundary extractive technique of particle cluster algorithm | |
| CN107180436A (en) | A kind of improved KAZE image matching algorithms | |
| CN106780508A (en) | A kind of GrabCut texture image segmenting methods based on Gabor transformation | |
| CN109166167B (en) | Multi-quality interface extraction method based on point set voxels | |
| CN113093216A (en) | Irregular object measurement method based on laser radar and camera fusion | |
| CN113177897A (en) | Rapid lossless filtering method for disordered 3D point cloud | |
| CN110363775A (en) | A kind of image partition method based on domain type variation level set | |
| CN114663373A (en) | Point cloud registration method and device for detecting surface quality of part | |
| Jin et al. | Defect identification of adhesive structure based on DCGAN and YOLOv5 | |
| CN106910180A (en) | A kind of image quality measure method and device | |
| CN104821015B (en) | Surface sampling point α-shape curved surface topology rebuilding methods in kind | |
| CN110955809B (en) | High-dimensional data visualization method supporting topology structure maintenance | |
| CN116452604B (en) | Complex substation scene segmentation method, device and storage medium | |
| CN108510591A (en) | A kind of improvement Poisson curve reestablishing method based on non-local mean and bilateral filtering | |
| Liao et al. | Fast hierarchical animated object decomposition using approximately invariant signature | |
| Lyu et al. | Laplacian-based 3D mesh simplification with feature preservation | |
| Sawdayee et al. | Orex: Object reconstruction from planar cross-sections using neural fields | |
| Wang et al. | A feature preserved mesh simplification algorithm | |
| CN115131384A (en) | Bionic robot 3D printing method, device and medium based on edge preservation | |
| CN108876711A (en) | A kind of sketch generation method, server and system based on image characteristic point |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| 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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20180202 Termination date: 20180713 |