CN108765478A - It is a kind of to build the density clustering algorithm that separating monomer is built in point cloud - Google Patents
It is a kind of to build the density clustering algorithm that separating monomer is built in point cloud Download PDFInfo
- Publication number
- CN108765478A CN108765478A CN201810573354.3A CN201810573354A CN108765478A CN 108765478 A CN108765478 A CN 108765478A CN 201810573354 A CN201810573354 A CN 201810573354A CN 108765478 A CN108765478 A CN 108765478A
- Authority
- CN
- China
- Prior art keywords
- point
- building
- point cloud
- built
- density
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/50—Depth or shape recovery
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- 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
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30181—Earth observation
Abstract
The present invention relates to the density clustering algorithms that separating monomer in a kind of building point cloud is built, and Density Clustering thought is applied to the separation process of single building.Antinoise of the present invention, robustness is good, calculated performance is excellent, high to the good accuracy of the data segmentation effect in building dense area, is not necessarily to manual intervention.The single building point cloud (data set) being automatically separated by this algorithm, it can be directly used for production building threedimensional model, and then it is used as data basis, wide range of services is built in smart city or digital city, is applied to the fields such as City Building data real-time update, city fining, scientific management or operation.
Description
Technical field
The present invention relates to computerized three-dimensional reconstruction, the Density Clustering that separating monomer is built in especially a kind of building point cloud is calculated
Method.
Background technology
The data source that three-dimensional model building is rebuild, it is another more efficient in addition to traditional artificial mapping, stereopsis
More accurate mode is using the high-precision three-dimensional point cloud (Point Cloud) acquired in laser radar (LiDAR).Based on a cloud
The key step for rebuilding building threedimensional model includes construction area extraction (filtering of point cloud and classification) and three-dimensional reconstruction.It is full
The demands such as sufficient theme inquiry, the three-dimensional building model of reconstruction should be physically differentiable.But work as construction zone
(building point cloud) after classifying (extraction) in original point cloud out, it includes often that more buildings are built that building point cloud data, which is concentrated,.In order to
Correctly, three-dimensional model building is effectively rebuild, needs therefrom to isolate each single building.Existing single building separation method
It can be divided into the separation method based on grid or based on vector by data processing method.Separation process based on grid will first put cloud
Data are projected in 2D planes by certain rule (such as height, dot density) resampling, are then used image procossing or correlation again
Technological means, cluster, segmentation, after the methods of edge detection, tracking isolate single building region, then back projection, and then divide
Go out single building point cloud.The main difficulty of Grid Method includes:Suitable cell size is selected, suitable segmentation threshold is selected
Deng in addition data conversion also brings along error, leads to incorrect segmentation result.Most common method is based on sky in vector method
Between domain cluster.Sampath and Shan (2007) are based on region growth method (or Moving split-window technique) cluster separating monomer and build
Object is built, core concept is to search for all adjoint points in its window neighborhood, then move since certain point using search window
New adjoint point (moving window, moving window) is added in the window's position.Awrangjeb and Fraser (2014) use DEM
Building mask (building mask) is generated, clusters the planar point in mask to form planar chip, remerges adjacent plane
Piece forms single building region.The boundaries for improving algorithm of convex hull tracking building point cloud Kwak and Habib (2014), will be away from
From the building point cloud segmentation open form more than threshold value at single building.Several main problems of the above method are:Noise-sensitive, parameter
(threshold value) is sensitive, and calculates complicated.In addition, aforementioned all methods (segmentation accuracy poor to the adaptability in building dense area
It is relatively low).
Invention content
The shortcomings that the invention solves the above-mentioned prior arts provides one kind and avoiding error caused by data conversion, reduces
The subjectivity and non-extensive adaptability of all kinds of threshold values (parameter) setting, antinoise, robustness are good, algorithm calculated performance it is excellent and
The density clustering algorithm that separating monomer is built in the building point cloud being easily achieved.
The present invention solves the technical solution of its technical problem use:The density that separating monomer is built in this building point cloud is poly-
Class algorithm, includes the following steps:
1) from any untreated point P0Start, if including at least minPts point, P in its ε-neighborhood0For core point,
Start to go to step 2) progress cluster process;Otherwise, by P0Labeled as noise spot, and repeat current procedures;If institute in data set
It is a little all disposed, then cluster process terminates;
2) by P0ε-neighborhood point be added in new cluster C, and by these neighborhood points according to its respective adjoint point quantity,
It is respectively labeled as core point or noise spot;
3) it repeats step 2) and continues same processing procedure that is, by all untreated core points in C, until not new
Core point be added C until, that is, find a density connection maximal subset, mark/isolate a certain single building;
4) it repeats the above steps, finds a new cluster.
Preferably, calculate neighborhood apart from when use Min Ke Paderewski distance measures, about n- tie up real number space in
Two points,Its p- rank Min Ke Paderewski distance definition is as follows:
Wherein 1,2 or ∞ can be used in the value of p.
Invention has the advantages that:Antinoise of the present invention, robustness is good, calculated performance is excellent, to the number in building dense area
It is high according to the good accuracy of segmentation effect, it is not necessarily to manual intervention;The single building point cloud (data set) being automatically separated by this algorithm, can
It being directly used in production building threedimensional model, and then as data basis, wide range of services is built in smart city or digital city,
Applied to fields such as City Building data real-time update, city fining, scientific management or operations.
This algorithm arithmetic speed is fast, and computational efficiency is high, and the accuracy of segmentation result is high;It effectively overcomes other and has list
Body builds the defect (sensitive, less efficient, calculating complexity of threshold value etc.) of separation algorithm, even more in the world for the first time by Density Clustering
Thought is applied to the separation process of single building.
Description of the drawings
Fig. 1 is the algorithm process flow diagram of the present invention;
Fig. 2 is p- rank Minkowski parasangs circle (2D):(a) p=1, (b) p=2, (c) p=∞, (d) three's overlapping
Comparison;
Fig. 3 is single building separation example (sample 1):(a) classification extraction construction area LiDAR point cloud, (b) monomer build
Separating resulting (different colours label) is built, (c) the reference image (the German regions Vaihingen sample data) of corresponding region;
Fig. 4 is Density Clustering process schematic:Since certain core point, a certain monomer is extracted after iteration several times and is built
It builds;
Fig. 5 is algorithm alleged by the present invention and Moving split-window technique (method alleged by bibliography 1) contrast test (sample 2):(a)
Sorted building point cloud, (b) the reference image of corresponding region, (c) segmentation result (each single building in figure of inventive algorithm
All correct segmentations), (d) Moving split-window technique separating resulting (two, right side building fails correctly to separate).
Specific implementation mode
The invention will be further described below in conjunction with the accompanying drawings:
Density clustering (density-based clustering) is printenv method, it is believed that cluster is about close
The high-density region of function p (x) is spent, known cluster numbers k is not necessarily to.Intuitively, density clustering is highdensity continuous
The data subset of (contiguous region) is divided by the continuum of low-density on region, is formed through such method
Cluster be also referred to as " natural cluster (natural cluster) ".If it is spatial data, then cluster result is more intuitive, such as quilt
River, the region etc. that road etc. is divided.Directly from building in the plane geometry distribution or natural form from the point of view of, building
Object is exactly each cluster naturally.When construction zone in original LiDAR point cloud after separating, include in data set
Several single buildings.Obviously, building is exactly "high" density region, and the non-construction zone (ground, vegetation etc.) rejected
Then form low density area.Under normal conditions, each single building point cloud data (high density area) can be surrounded by density regions.It is close
Degree cluster is defined as follows:
Density Clustering:A sampling point set on real number space is tieed up to given d,Its data obeys certain
The unknown probability distribution of kind, density function are p (x), then cluster is the high density subset on D about p (x), that is, close to what is given
Degree threshold value λ, all high density object sets x | p (x)>λ }, the density collection on referred to as threshold level λ.
During Density Clustering, it is (opposite that cluster (or data group) is defined as the high density continuum on data space
In other regions of data set), (point of non-tight distribution needs to isolate in cluster process the object in density regions
Come) it is taken as noise (outlier).
This algorithm can be automatically separated the single building of any shape, and cluster density threshold level therein is by giving
" distance " threshold epsilon and neighborhood in adjoint point number minPts determined by, that is, given by " distance " function and threshold value institute
The density estimation of point in determining contiguous range.Algorithm process flow is following (with reference to figure 1):
1) from any untreated point P0Start, if including at least minPts point in its ε-neighborhood, starts to cluster
Journey is (that is, P0It for core point, goes to step 2));Otherwise, by P0Labeled as noise spot (outlier), and repeat current procedures;If
All the points are all disposed in data set, then cluster process terminates;
2) by P0ε-neighborhood point be added in new cluster C, and by these neighborhood points according to its respective adjoint point quantity,
It is respectively labeled as core point or noise spot;
3) it repeats step 2) and continues same processing procedure that is, by all untreated core points in C, until not new
Core point be added C until (find a density connection maximal subset, that is, mark/isolate a certain single building);
4) it repeats the above steps, finds a new cluster.
When being built using density-based spatial clustering algorithm separating monomer, it is related to two parameters, i.e. neighborhood distance threshold ε
And the minPts that at least counts.Judge whether a point is effective cluster point, depends in its ε-neighborhood whether including enough
It counts (>=minPts).Wherein, ε is related with problem to be solved, and minPts is then the minimum capacity of desired cluster.?
In given data set, ε is smaller, and it is also smaller to be formed by cluster;Or in other words, ε is bigger, then clusters also bigger.To airborne
For LiDAR point cloud data, the lower limit of ε is a cloud scanning line direction and scan line spacings direction equalization point away from larger in the two
Person.For spatial data, minimum adjoint point number minPts is then related with Spatial Dimension D, that is, minPts >=D+1 (if being less than D,
Without practical significance).For the data set of Noise, relatively large minPts values will produce relatively good cluster/segmentation knot
Fruit.
Distance function (or dissmilarity degree is estimated) also influences the selection of distance threshold ε and minimum adjoint point number minPts.This calculation
Method can be applied to a variety of different distance measures, e.g., physical distance (Euclidean distance), rectangular distance (block according to actual demand
Distance, chessboard distance) neighborhood etc..Or directly it is summarised as Minkowski Distance (Minkowski distance).About n-
Two points in real number space are tieed up,Its p- rank Minkowski distance definition is as follows:
The common values of p are 1,2 and ∞, correspond to city block distance (Manhattan distance), Euclidean distance and chess respectively
Disk distance (Chebyshev distance)." unit circle " (2D) of these three distances is as shown in Figure 2.It can be seen from the figure that
The unit circle of the p of higher order includes the unit circle of low order p, that is, given ε, by Chebyshev apart from identified neighborhood " area "
Maximum may include more points in neighborhood.
From the point of view of data distribution, minPts is the smoothing factor in density estimation.It is unknown in data distribution, and give
Under conditions of minPts (desired min cluster), ε appropriate can be selected away from figure (k-distance graph) by k-.
Construction process of the k- away from figure be, analyzes the distance that each point in data set arrives a closest approach of its kth-, and with this apart from descending row
Sequence forms a curve.It indicates to start to change to core point from noise spot at curve first " inflection point ", i.e. the left part
Neighborhood is counted<K, and its right part neighborhood points >=k.Select the corresponding distance in the place as ε-neighborhood distance threshold i.e.
It can.
Below using the sample data in the German regions Vaihingen as embodiment explanation.Three-dimensional laser point cloud data is collected in
2010, the average dot density for selecting cloud was 6 points/rice2.Building point cloud data (collection) after this algorithm only application class, wherein sample
1 data of example as with shown in 3.Subgraph (a) is the construction area laser point cloud that is extracted, it can be seen that construction area (grey)
It is surrounded by non-building area domain (the non-building point filtered out, white), that is, corresponding high-density region and low-density in Density Clustering
Region;Subgraph (b) is single building separating resulting, wherein each single building is distinguished in different colors;Subgraph (c) corresponding region
With reference to image.
Fig. 4 provides Density Clustering and detaches the interim cluster process signal of certain single building, that is, is opened from seed point (core point)
Begin, seed queue is added in the abutment points for meeting density threshold condition in each iterative process, continues iterative process, to separation
Until going out the single building.
Fig. 5 is algorithm alleged by the present invention (density clustering algorithm that separating monomer is built in building point cloud) and existing movement
Windowhood method (method alleged by bibliography 1) contrast test result (sample 2):Wherein, (a) is sorted building point cloud data
Collection, is used as the data input of single building separation test, is (b) the reference image of corresponding region, for intuitively observing data field
Domain, (c) be inventive algorithm segmentation result (in figure with different colours labeled monomer build segmentation result), (d) be Moving Window
The separating resulting of mouth method;It can be seen from the figure that (algorithm) of the invention is correctly partitioned into all single buildings, but moving window
Method fails correctly to distinguish two buildings on right side.
In addition to the implementation, the present invention can also have other embodiment.It is all to use equivalent substitution or equivalent transformation shape
At technical solution, fall within the scope of protection required by the present invention.
Claims (2)
1. the density clustering algorithm that separating monomer is built in a kind of building point cloud, includes the following steps:
1) from any untreated point P0Start, if including at least minPts point, P in its ε-neighborhood0For core point, start
Go to step 2) progress cluster process;Otherwise, by P0Labeled as noise spot, and repeat current procedures;If all the points in data set
It is all disposed, then cluster process terminates;
2) by P0ε-neighborhood point be added in new cluster C, and by these neighborhood points according to its respective adjoint point quantity, respectively
Labeled as core point or noise spot;
3) it repeats step 2) and continues same processing procedure that is, by all untreated core points in C, until not new core
Until C is added in heart point, that is, the maximal subset of density connection is found, a certain single building is marked/isolate;
4) it repeats the above steps, finds a new cluster.
2. the density clustering algorithm that separating monomer is built in building point cloud according to claim 1, it is characterized in that:It is calculating
Neighborhood apart from when use Min Ke Paderewski distance measures, about n- tie up real number space in two points,Its p- rank
Min Ke Paderewski distance definitions are as follows:
Wherein 1,2 or ∞ can be used in the value of p.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810573354.3A CN108765478A (en) | 2018-06-06 | 2018-06-06 | It is a kind of to build the density clustering algorithm that separating monomer is built in point cloud |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810573354.3A CN108765478A (en) | 2018-06-06 | 2018-06-06 | It is a kind of to build the density clustering algorithm that separating monomer is built in point cloud |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108765478A true CN108765478A (en) | 2018-11-06 |
Family
ID=63999201
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810573354.3A Pending CN108765478A (en) | 2018-06-06 | 2018-06-06 | It is a kind of to build the density clustering algorithm that separating monomer is built in point cloud |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108765478A (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109489580A (en) * | 2018-12-10 | 2019-03-19 | 华东理工大学 | A kind of processing of complex surface in machine point cloud detection and compensation method |
CN109613553A (en) * | 2018-12-18 | 2019-04-12 | 歌尔股份有限公司 | The method, apparatus and system of physical quantities in scene are determined based on laser radar |
CN111340822A (en) * | 2020-02-24 | 2020-06-26 | 武汉大学 | Multi-scale self-adaptive airborne LiDAR point cloud building single segmentation method |
CN111402415A (en) * | 2020-03-12 | 2020-07-10 | 腾讯科技(深圳)有限公司 | Object body elevation map generation method and device, storage medium and terminal equipment |
WO2021087760A1 (en) * | 2019-11-05 | 2021-05-14 | 深圳市大疆创新科技有限公司 | Target detection method, radar, device, and storage medium |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102289585A (en) * | 2011-08-15 | 2011-12-21 | 重庆大学 | Real-time monitoring method for energy consumption of public building based on data mining |
-
2018
- 2018-06-06 CN CN201810573354.3A patent/CN108765478A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102289585A (en) * | 2011-08-15 | 2011-12-21 | 重庆大学 | Real-time monitoring method for energy consumption of public building based on data mining |
CN102289585B (en) * | 2011-08-15 | 2014-06-18 | 重庆大学 | Real-time monitoring method for energy consumption of public building based on data mining |
Non-Patent Citations (2)
Title |
---|
刘士程: "基于LiDAR数据提取建筑物顶面轮廓线方法研究", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
程德强 著者: "《流媒体覆盖网络及其关键技术研究》", 31 October 2008 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109489580A (en) * | 2018-12-10 | 2019-03-19 | 华东理工大学 | A kind of processing of complex surface in machine point cloud detection and compensation method |
CN109489580B (en) * | 2018-12-10 | 2021-09-24 | 华东理工大学 | On-machine point cloud detection and compensation method for surface processing of aero-engine blade |
CN109613553A (en) * | 2018-12-18 | 2019-04-12 | 歌尔股份有限公司 | The method, apparatus and system of physical quantities in scene are determined based on laser radar |
WO2021087760A1 (en) * | 2019-11-05 | 2021-05-14 | 深圳市大疆创新科技有限公司 | Target detection method, radar, device, and storage medium |
CN111340822A (en) * | 2020-02-24 | 2020-06-26 | 武汉大学 | Multi-scale self-adaptive airborne LiDAR point cloud building single segmentation method |
CN111402415A (en) * | 2020-03-12 | 2020-07-10 | 腾讯科技(深圳)有限公司 | Object body elevation map generation method and device, storage medium and terminal equipment |
CN111402415B (en) * | 2020-03-12 | 2021-06-01 | 腾讯科技(深圳)有限公司 | Object body elevation map generation method and device, storage medium and terminal equipment |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108765478A (en) | It is a kind of to build the density clustering algorithm that separating monomer is built in point cloud | |
CN107146280B (en) | Point cloud building reconstruction method based on segmentation | |
CN112489212B (en) | Intelligent building three-dimensional mapping method based on multi-source remote sensing data | |
CN112070769B (en) | Layered point cloud segmentation method based on DBSCAN | |
CN106803253B (en) | A kind of three-dimensional rock image crack identification method | |
CN109118500B (en) | Image-based three-dimensional laser scanning point cloud data segmentation method | |
CN110599506B (en) | Point cloud segmentation method for three-dimensional measurement of complex special-shaped curved surface robot | |
JP2017521779A (en) | Detection of nuclear edges using image analysis | |
CN110992381A (en) | Moving target background segmentation method based on improved Vibe + algorithm | |
CN110222642A (en) | A kind of planar architectural component point cloud contour extraction method based on global figure cluster | |
CN105046694A (en) | Quick point cloud registration method based on curved surface fitting coefficient features | |
CN115205690B (en) | Method and device for extracting street tree in monomer mode based on MLS point cloud data | |
CN108596920A (en) | A kind of Target Segmentation method and device based on coloured image | |
CN113484875B (en) | Laser radar point cloud target hierarchical identification method based on mixed Gaussian ordering | |
CN108765409A (en) | A kind of screening technique of the candidate nodule based on CT images | |
CN116258857A (en) | Outdoor tree-oriented laser point cloud segmentation and extraction method | |
CN115661072A (en) | Disc rake surface defect detection method based on improved fast RCNN algorithm | |
CN109359680B (en) | Explosion sillar automatic identification and lumpiness feature extracting method and device | |
CN114140485A (en) | Method and system for generating cutting track of main root of panax notoginseng | |
CN113409332A (en) | Building plane segmentation method based on three-dimensional point cloud | |
CN108520550A (en) | Point cloud data Poisson curve reestablishing method based on noise classification and MLS | |
CN110348478B (en) | Method for extracting trees in outdoor point cloud scene based on shape classification and combination | |
CN105374024B (en) | The method of high-resolution satellite image on-water bridge extraction | |
CN114140466B (en) | Plant root system measuring method, system and device based on image processing | |
CN115170950A (en) | Outdoor scene building extraction method based on multi-feature constraint |
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 | ||
WD01 | Invention patent application deemed withdrawn after publication | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20181106 |