CN104573705A - Clustering method for building laser scan point cloud data - Google Patents
Clustering method for building laser scan point cloud data Download PDFInfo
- Publication number
- CN104573705A CN104573705A CN201410539285.6A CN201410539285A CN104573705A CN 104573705 A CN104573705 A CN 104573705A CN 201410539285 A CN201410539285 A CN 201410539285A CN 104573705 A CN104573705 A CN 104573705A
- Authority
- CN
- China
- Prior art keywords
- point
- circle
- minpts
- cloud
- value
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Probability & Statistics with Applications (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a clustering method for building laser scan point cloud data. The method comprises the following steps: converting the point cloud data into two-dimensional data to acquire a set X; appointing density threshold value MinPts; calculating the longest distance of MinPts objects with shortest distance from the point towards any point in the X; counting maximal and minimum value of the longest distance of all the points; classifying the difference value of the maximal and minimum value of the longest distance into n equal portions; making a circle by adopting the point generating the minimum value of the longest distance as the center and the gradual increasing value of the distance of the equal portions as the radius; calculating the number of points in each circle; adopting the difference value as a abscissa; adopting the numbers of the points in each circle as an ordinate; conducting fitting to form a curve; seeking points of inflexion of the curve; taking the number value of the abscissa corresponding to the points of inflexion as the value of sigma; building AQ-DBSCAN algorithm by adopting the MInPts and the sigma as conditions; conducting clustering on the points in the X to obtain the belonged cluster analysis of the points in the point cloud in the parts of the building.
Description
Technical field
The present invention relates to a kind of clustering method of buildings Point Cloud of Laser Scanner.
Background technology
Ancient building is the important symbol of human civilization, is the special carrier of cultural information.Therefore, inherit and protect ancient building to be contemporary people inescapable responsibility.But due to current expanding economy and urban construction, and ancient building was corroded by the years, make a lot of Large-scale Ancient build the deformation and failure created in various degree, conservation of historic buildings is faced with severe situation.
Three-dimensional color scanning technology has rapidity; not contact; penetrability; in real time, dynamically, initiative; high density, high precision; the characteristic such as digitizing, robotization, along with the progress of aspect and the reductions of price such as its measuring accuracy, sweep velocity, spatial resolution, obtains applying more and more widely in conservation of historic buildings.It adopts non-contact measurement means, can when not damaging object, be deep into complex environment and scan operation is carried out at scene, and directly various three-dimensional data that is large-scale, complicated, irregular entity intactly collected in computing machine, thus quick reconfiguration goes out the three-dimensional model of scanning object.Meanwhile, the three-dimensional laser point cloud data that it gathers not only comprises the spatial information of target, and have recorded Reflection intensity information and the color half-tone information of target.
Utilize three-dimensional laser scanner can obtain the three-dimensional point cloud model of building fast, each point in cloud data forgives dominant three-dimensional coordinate.Based on cloud data, simple buildings three dimensional navigation and roaming can be realized.But, cloud data structured data amount simple is on the one hand huge, be difficult to be applicable to the efficient roaming of heritage buildings and browse, on the other hand due to the information of semantic (semantic) rank cannot be provided, cannot higher level analysis and research be carried out, therefore need to process cloud data further on this basis, thus extract and rebuild the three-dimensional structure of object, and to realize display, analysis, measurement, emulation, simulation, monitoring, store, the functions such as retrieval.And Point Cloud Processing needs man-machine interaction at present to a great extent at present, automaticity is lower.
Data Segmentation is the important step of cloud data feature extraction and three-dimensional modeling.Rabbani et al (2006) data of description is divided into the pointwise identification procedure of cloud data, and the point with like-identified is considered to belong to same surface or region, and those points in continuum with similar features are split to a subset.
Current some cloud partitioning algorithm mainly comprises four classes: based on the segmentation of cluster, the segmentation increased based on region, based on the segmentation of models fitting and other mixing partitioning algorithms.
General clustering algorithm roughly comprises five kinds: the method for hierarchical method, division methods, density based, the method based on model and the method based on grid.
Density clustering method is thought, bunch is by the high density subject area that density regions separates in data space, and the data in sparse data region are considered to noise data.This algorithm sets certain threshold values, as long as the density of certain some adjacent domain is greater than this threshold value, cluster just can proceed.These class methods can find the cluster of arbitrary shape, and can filter " noise " data.What can find arbitrary shape bunch is the great advantage of this algorithm, and its major defect is that more responsive to user-defined density parameter, different threshold values is larger for the Influence on test result of cluster.The typical algorithm of these class methods comprises DBSCAN algorithm (Density-Based Spatial Clustering of Applications with Noise) etc.
Summary of the invention
Because the cluster of the 2-D data of heritage buildings cloud data in Gaussian sphere or other curves has various shape, and wherein there is a large amount of noise points.Therefore density clustering Measures compare is applicable to the cluster of such data.For the feature of ancient building cloud data, the present invention proposes the AQ-DBSCAN algorithm based on DBSCAN algorithm improvement, and provide a kind of AQ-DBSCAN algorithm that utilizes to the clustering method of buildings Point Cloud of Laser Scanner, AQ-DBSCAN algorithm of the present invention solves the automatic estimation problem of the radius of neighbourhood in DBSCAN algorithm, and reduces operand to carry out the problem of Density Clustering process fast when effectively spreading in Selecting Representative Points from A and cluster.
AQ-DBSCAN algorithm application in the cluster of ancient building cloud data, is belonged to which position in buildings with the point obtained in a cloud by the present invention, and for this reason, technical scheme provided by the invention is:
A clustering method for buildings Point Cloud of Laser Scanner, comprising:
Step one, each point in the three dimensional point cloud obtained after carrying out laser scanning to buildings is converted into 2-D data, obtains the set X of the 2-D data of some cloud;
Step 2, specified density threshold value is minimum comprises the MinPts that counts, for any one point in set X, calculate and the nearest minimum maximum distance comprising MinPts object of counting of this point, and statistics to gather in X the maximal value of maximum distance a little and minimum value;
Step 3, the maximal value of maximum distance and the difference m of minimum value are divided into n equal portions, to produce the point of the minimum value of maximum distance in a cloud for the center of circle, in units of the spacing m/n of equal portions, increment value is radius step by step, makes n circle, calculates the quantity of point in each circle;
Step 4, using difference m as horizontal ordinate, interval as the incremental units of horizontal ordinate using the spacing m/n of equal portions, with the quantity N put in each circle
mfor ordinate, draw coordinate diagram, and the point on Fitting Coordinate System figure, form curve plotting;
The point of inflexion on a curve that step 5, searching are drawn, using the value of the numerical value of horizontal ordinate corresponding for this flex point as radius of neighbourhood σ;
Step 6, with density threshold MinPts and radius of neighbourhood σ for condition sets up AQ-DBSCAN algorithm, and with AQ-DBSCAN algorithm to set X in point carry out cluster, belong to the cluster analysis at which position in this buildings with the point obtained in described some cloud.
Preferably, in the clustering method of described buildings Point Cloud of Laser Scanner, described step 6 comprises:
6.1) to produce the point of the minimum value of maximum distance in a cloud for the center of circle, with radius of neighbourhood σ for radius, draw first order circle, if the number of the point in this first order circle is less than MinPts, then eliminate this first order circle, if be greater than MinPts; Then continue 6.2)
6.2) with c σ wherein 0 < c < 1 for radius, draw first order circle, if the first order is round and the number of point in annular region between the sub-circle of the first order is less than MinPts, then choose points all in annular region as the center of circle, the second level; If the number of the point in annular region is more than or equal to MinPts, then choose MinPts point as the center of circle, the second level;
6.3) with the selected center of circle, the second level for the center of circle, with radius of neighbourhood σ for radius, draw at least one second level circle, if the number of point in the circle of any one second level is less than MinPts, then eliminate this second level circle, if be greater than MinPts; Then repeated execution of steps 6.2) and 6.3), draw circle step by step, until the circle drawn all is eliminated;
6.4) being gathered by the point in the circle of all ranks is a class, and removes in described some cloud, and to the operation of the some repeated execution of steps two in remaining some cloud to step 6.
Preferably, in the clustering method of described buildings Point Cloud of Laser Scanner, described step 6.2) in, choosing MinPts the method put as the center of circle, the second level is:
In this annular region, first choose from first order distance of center circle from farthest o'clock as first o'clock, then this distance of first of selected distance point is farthest as second point, then the distance sum of selected distance this first and second point point is farthest as thirdly, according to this rule until select MinPts point.
Preferably, in the clustering method of described buildings Point Cloud of Laser Scanner, described MinPts is 4.
Preferably, in the clustering method of described buildings Point Cloud of Laser Scanner, described c is 3/4.
Preferably, in the clustering method of described buildings Point Cloud of Laser Scanner, described flex point is the maximum point of slope variation.
Preferably, in the clustering method of described buildings Point Cloud of Laser Scanner, in described step one, each point in the three dimensional point cloud obtained after carrying out laser scanning to buildings is mapped to Gaussian sphere thus is converted into 2-D data.
Introduce the key concept of DBSCAN algorithm below.
DBSCAN algorithm:
1 x in point set X
ithe density at place generally can be defined as follows:
Wherein U (x
i) be an x
iweight;
be kernel function, kernel function selects the symmetrical density function at initial point usually, as Gaussian function, Epanechnikov kernel function etc.; σ is the bandwidth of kernel function, actually represent the r ball neighborhood that take σ as radius.Order
U(x
i)＝{1|x
i∈X}
Then f
dx () is exactly the definition to density in DBSCAN algorithm.
Define 1 σ neighborhood: data-oriented set X, x
0an object in set X, then with x
0centered by, be that the dimension suprasphere region of radius is called x with σ
0σ neighborhood σ x
0, that is:
σ(x
0)＝{x∈X|D(x，x
0)≤σ}
Wherein, D (x, x
0) represent x and x
0between distance.
Define 2 core points: for x
0∈ X, given integer MinPts, if σ is (x
0) in object number meet | N σ (x
0) |>=MinPts, then claim x
0for the core point under (σ, MinPts) condition.Drop in the σ neighborhood of certain core point, but be not the object of core point, be called frontier point.
Definition 3 direct density can reach: under condition (σ, MinPts), if object x and x
0meet:
x∈σ(x
0)；
|Nσ(x
0)|≥MinPts。
Then x is claimed to be from x
0direct density can reach.
Define 4 density can reach: in data set X, for object sequence x
0, x
1, x
2..., x
nif, under condition (σ, MinPts), x
ito x
i+1be (0≤i < n) that direct density can reach, then claim object x
0to x
nthat density can reach.
Define 5 density to be connected: data-oriented collection X, under (σ, MinPts) condition, if there is object x
0, x
iand x
j, make x
0to x
iand x
0to x
jbe that density can reach, so claim x
iand x
jthat density is connected under (σ, MinPts) condition.
Define 6 bunches with noise: the object that all density is connected forms one bunch, if an object is not in any bunch, so this object is called noise.
The algorithm steps of DBSCAN is specific as follows:
DBSCAN algorithm needs these two parameters of σ and MinPts given in advance, and the point that directly can be reached by iterative search density is set up bunch.
The geometric configuration of heritage buildings component mainly comprises the type such as plane, cylinder, and being mapped in Gaussian sphere is the shapes such as point, line.DBSCAN algorithm can identify the cluster of arbitrary shape, and the primary segmentation for heritage buildings Gaussian mapping point cloud has stronger applicability.Because DBSCAN algorithm will carry out Region Queries and calculating for each point, its operation efficiency is lower, and the time complexity of DBSCAN algorithm is O (n
2).For the very large ancient building point cloud of data volume, the efficiency of DBSCAN clustering algorithm especially seems low.
DBSCAN algorithm, when determining (σ, MinPts) parameter, needs to carry out statistical computation, draws the curve map of density and some number, then determines parameter by manually choosing, and manually chooses obvious inapplicable robotization ancient building point cloud segmentation demand.
DBSCAN algorithm needs when the growth of certain classification to judge whether each point in neighborhood is core point, this will calculate have a σ (x
i) some number in neighborhood.Although the cloud data after Gaussian mapping can set up MultyGrid-KD tree index to accelerate judgement speed, the operational efficiency of this algorithm is still lower.
The FDBSCAN algorithm that Zhou Shuigeng etc. (2000) propose is the innovatory algorithm for DBSCAN.FDBSCAN algorithm proposes to select to specify the representative point (instead of all neighborhoods point) of number to replace carrying out classification growth in neighborhood.The number of representative point is relevant with the dimension in space, and namely the representative point number of n-dimensional space is 2n, and namely representative point here refers to second level round dot, third level round dot etc. except round dot.
As shown in Figure 2, the diversity of two-dimensional space four representative points is illustrated on the impact of row cluster expansion effect.When the diversity of representative point is bad, there is certain point and x
idensity can reach, but searches for the situation less than this point by representative point neighborhood search.Fig. 2 mid point A points to x
i, 1 circle pointed to is σ (x
i) regional extent, some B be from σ (x
i) representative point choosing out, i.e. the center of circle, the second level, 2 circles pointed to are the neighborhood regional extents in representative point and the center of circle, the second level, and some C is the representative point being used for carrying out cluster expansion and the third level center of circle elected further from the neighborhood of representative point B.In this figure, because representative point all concentrates on the upper area of neighborhood, cause the propagation direction of cluster to develop all upward, below point object cannot be incorporated into in identical category.
For above problem, need on the one hand to carry out merging treatment in the follow-up relevant group to these points and formation.On the other hand on representative point is chosen, the rapidity of algorithm and the diffusivity of representative point be taken into account.Zhou Shuigeng etc. (2000) propose the Algorithms of Selecting of two representative points, but are all not suitable for the intensive cloud data in Gaussian sphere on the diversity of efficiency of algorithm and representative point.
Relative to DBSCAN, FDBSCAN algorithm, the improvement of AQ-DBSCAN algorithm is mainly reflected in 2 points: one is under the prerequisite of given MinPts parameter, and provide the automatic estimation of σ, two is achieve density clustering algorithm more fast.
Relative to FDBSCAN algorithm, AQ-DBSCAN is in the two-dimensional characteristics of searching algorithm except foundation Gaussian sphere Surface Data, representative point number is decided to be 4, thus outside minimizing operand, also proposed the representative point Algorithms of Selecting being applicable to intensive cloud data, representative point is improved on the guarantee diffusible basis of representative point and chooses efficiency.
Accompanying drawing explanation
Fig. 1 is using difference m as horizontal ordinate in the present invention, with the quantity N put in each circle
mfor the curve plotting that vertical seat matching is formed;
Fig. 2 is that the diversity of two-dimensional space four representative points in FDBSCAN algorithm is to the effect diagram of row cluster expansion effect;
Fig. 3 (a) is the cloud data figure in an embodiment in the present invention;
Fig. 3 (b) is the Gaussian mapping figure of the cloud data corresponding with Fig. 3 (a);
Fig. 3 (c) is the N of programming count in one embodiment of the invention
mcurve map;
Fig. 3 (d) be utilize AQ-DBSCAN algorithm to obtain in one embodiment of the invention Gaussian sphere on Clustering Effect figure;
Fig. 3 (e) is the Clustering Effect figure corresponded in one embodiment of the invention on space curved surface;
Fig. 3 (f) be utilize DBSCAN algorithm to obtain in one embodiment of the invention Gaussian sphere on Clustering Effect figure;
Fig. 3 (g) carries out the segmentation result figure to the outer pillar cloud data of the Hall of Supreme Harmony after next step process again after utilizing AQ-DBSCAN algorithm in one embodiment of the invention;
Fig. 4 (a) is the cloud data figure in another embodiment in the present invention;
Fig. 4 (b) is the Gaussian mapping figure of the cloud data corresponding with Fig. 4 (a);
Fig. 4 (c) is the N of programming count in another embodiment of the present invention
mcurve map;
Fig. 4 (d) be utilize AQ-DBSCAN algorithm to obtain in another embodiment of the present invention Gaussian sphere on Clustering Effect figure;
Fig. 4 (e) is the Clustering Effect figure corresponded in another embodiment of the present invention on space curved surface;
Fig. 4 (f) be utilize DBSCAN algorithm to obtain in another embodiment of the present invention Gaussian sphere on Clustering Effect figure;
Fig. 4 (g) carries out the segmentation result figure to the Hall of Supreme Harmony external door beam cloud data after next step process again after utilizing AQ-DBSCAN algorithm in another embodiment of the present invention;
Fig. 5 (a) is the cloud data figure in another embodiment in the present invention;
Fig. 5 (b) is the Gaussian mapping figure of the cloud data corresponding with Fig. 5 (a);
Fig. 5 (c) is the N of programming count in another embodiment of the present invention
mcurve map;
Fig. 5 (d) be utilize AQ-DBSCAN algorithm to obtain in another embodiment of the present invention Gaussian sphere on Clustering Effect figure;
Fig. 5 (e) is the Clustering Effect figure corresponded in another embodiment of the present invention on space curved surface;
Fig. 5 (f) be utilize DBSCAN algorithm to obtain in another embodiment of the present invention Gaussian sphere on Clustering Effect figure;
Fig. 5 (g) carries out the segmentation result figure to The Gate of Supreme Harmony part cloud data after next step process again after utilizing AQ-DBSCAN algorithm in another embodiment of the present invention;
Fig. 6 is the algorithm comparison diagram consuming time of DBSCAN and AQ-DBSCAN of the present invention;
Fig. 7 is the algorithm comparison diagram consuming time of FDBSCAN and AQ-DBSCAN of the present invention.
Embodiment
Below in conjunction with accompanying drawing, the present invention is described in further detail, can implement according to this with reference to instructions word to make those skilled in the art.
As shown in Figure 1, the invention provides a kind of clustering method of buildings Point Cloud of Laser Scanner, comprising:
Step one, each point in the three dimensional point cloud obtained after carrying out laser scanning is mapped in Gaussian sphere is converted into 2-D data to buildings, obtain the set X of the 2-D data of some cloud;
Step 2, specified density threshold value is minimum comprises the MinPts that counts, for any one point in set X, calculate and the nearest minimum maximum distance comprising MinPts object of counting of this point, and statistics to gather in X the maximal value of maximum distance a little and minimum value; In the division of 2-D data, generally get MinPts is 4.
Step 3, the maximal value of maximum distance and the difference m of minimum value are divided into n equal portions, to produce the point of the minimum value of maximum distance in a cloud for the center of circle, in units of the spacing m/n of equal portions, increment value is radius step by step, makes n circle, calculates the quantity of point in each circle;
Step 4, using difference m as horizontal ordinate, interval as the incremental units of horizontal ordinate using the spacing m/n of equal portions, with the quantity N put in each circle
mfor ordinate, draw coordinate diagram, and the point on Fitting Coordinate System figure, form curve plotting as shown in Figure 1; In Fig. 1, because plane and the cylinder maximum distance in Gaussian sphere has certain difference, therefore this curve shows heaving of the sea slightly in the process upwards extended.
Step 5, find the point of inflexion on a curve of drawing, this flex point is the maximum point of slope variation, using the value of the numerical value of horizontal ordinate corresponding for this flex point as radius of neighbourhood σ.The method finding flex point is: connect the starting point in curve as shown in Figure 1 and terminal formation straight line L, calculate the distance H of the every bit on this curve to straight line L, the point that statistics H is maximum, be this flex point.
The algorithm steps of the value programming count of radius of neighbourhood σ is specific as follows:
Step 6, with density threshold MinPts and radius of neighbourhood σ for condition sets up AQ-DBSCAN algorithm, and with AQ-DBSCAN algorithm to set X in point carry out cluster, belong to the cluster analysis at which position in this buildings with the point obtained in described some cloud.
Comprise in this step 6:
6.1) to produce the point of the minimum value of maximum distance in a cloud for the center of circle, with radius of neighbourhood σ for radius, draw first order circle, if the number of the point in this first order circle is less than MinPts, then eliminate this first order circle, if be greater than MinPts; Then continue 6.2), namely Selecting Representative Points from A carries out dispersing a little.
6.2) with c σ wherein 0 < c < 1 for radius, draw first order circle, if the first order is round and the number of point in annular region between the sub-circle of the first order is less than MinPts, then choose points all in annular region as the center of circle, the second level; If the number of the point in annular region is more than or equal to MinPts, then choose MinPts point as the center of circle, the second level.Wherein, choosing MinPts the method put as the center of circle, the second level is:
In this annular region, first choose from first order distance of center circle from farthest o'clock as first o'clock, then this distance of first of selected distance point is farthest as second point, then the distance sum of selected distance this first and second point point is farthest as thirdly, according to this rule until select MinPts point.
Namely x is established
0for the object of in point set X, make 0 < c < σ, then
c(x
0)＝{x∈σ(x
0)|ε＜D(χ，x
0)≤σ}
C (x
0) be x
0candidate's point set of the center of circle, the second level (i.e. representative point) in neighborhood.C (x
0) be with x
0centered by, the annular section between c and σ.At σ (x
0) outer peripheral areas choose the center of circle, the second level (representative point), contribute to strengthen representative diffusivity and reduce classification expansion time Region Queries frequency.C is more close to σ, then the quantity of candidate point is fewer, and the efficiency of search representative point is higher, but diversity a little also may be caused bad simultaneously.For ancient building cloud data feature, consider diffusivity and high efficiency factor, the c value of this algorithm is 3 σ/4.
6.3) with the selected center of circle, the second level for the center of circle, with radius of neighbourhood σ for radius, draw at least one second level circle, if the number of point in the circle of any one second level is less than MinPts, then eliminate this second level circle, if be greater than MinPts; Then repeated execution of steps 6.2) and 6.3), draw circle step by step, until the circle drawn all is eliminated;
6.4) being gathered by the point in the circle of all ranks is a class, and removes in described some cloud, and to the operation of the some repeated execution of steps two in remaining some cloud to step 6.
In the clustering method of described buildings Point Cloud of Laser Scanner, described step 6.2) in, choosing MinPts the method put as the center of circle, the second level is:
In this annular region, first choose from first order distance of center circle from farthest o'clock as first o'clock P1, then this distance of first of selected distance point is farthest as second point, then the distance sum of selected distance this first and second point point is farthest as thirdly, according to this rule until select MinPts point.
That is: set and existing represent point set as P, representative point to be searched is p
k, then
Wherein
Below the specific descriptions of the center of circle, second level Algorithms of Selecting:
The complete algorithm of AQ-DBSCAN of the present invention is as follows:
Enumerate the embodiment of the cluster of the some cloud of the buildings that some utilize method of the present invention to obtain below.The point cloud that the terrestrial Laser scanner data from the Forbidden City The Gate of Supreme Harmony and the Hall of Supreme Harmony in the present invention obtains.
In one embodiment, as shown in Figure 3, the present invention acquires the pillar data outside the Hall of Supreme Harmony and has carried out cluster.Fig. 3 (a) is the cloud data of the outer pillar of the Hall of Supreme Harmony.Counting of this Points Sample is 141999.Fig. 3 (b) is the Gaussian mapping figure corresponding with this pillar.Fig. 3 (c) is the N of AQ-DBSCAN algorithm programming count
mcurve map and the σ parameter value (0.067831) estimated.DBSCAN and FDBSCAN algorithm is main by artificial judgment N on parameter estimation
mthe shape of curve and the change point position of Yan Sheng trend.As can be seen from Fig. 3 (c), AQ-DBSCAN algorithm provides the result that σ value meets artificial judgment.Fig. 3 (d) is the Clustering Effect figure in Gaussian sphere, and because cylinder and part satellite thing have overlap in Gaussian sphere, therefore being gathered is a class, and these will be analyzed by later overlay region and split further.Fig. 3 (e) is the Clustering Effect figure corresponded on space curved surface.Fig. 3 (f) is the cluster adopting existing DBSCAN algorithm to obtain in Gaussian sphere, and as can be seen from Gaussian sphere, the Clustering Effect utilizing AQ-DBSCAN algorithm and DBSCAN algorithm to obtain is completely the same.Fig. 3 (g) carries out the segmentation result figure to the outer pillar point cloud of the Hall of Supreme Harmony after next step process again after utilizing method of the present invention, can find out, after utilizing method of the present invention, can obtain good applicability to the Data Segmentation of ancient building point cloud.
In another embodiment, as shown in Figure 4, the present invention acquires the data of the Hall of Supreme Harmony external door beam and has carried out cluster.Fig. 4 (a) is the cloud data of the Hall of Supreme Harmony outer pillar top crossbeam.Counting of this Points Sample is 642984.Fig. 4 (b) is Gaussian mapping figure corresponding with it.Fig. 4 (c) is the N of AQ-DBSCAN algorithm programming count
mcurve map and the σ parameter value (0.011236) estimated.As can be seen from Fig. 3 (c), AQ-DBSCAN algorithm provides the result that σ value meets artificial judgment.Fig. 4 (d) is 4 (a) Clustering Effect figure in Gaussian sphere, Fig. 4 (e) is the Clustering Effect figure corresponded on space curved surface, as can be seen from Fig. 4 (e), the front of crossbeam and side make a distinction by AQ-DBSCAN cluster.Fig. 4 (f) is the cluster adopting existing DBSCAN algorithm to obtain in Gaussian sphere, and as can be seen from Gaussian sphere, the Clustering Effect utilizing AQ-DBSCAN algorithm and DBSCAN algorithm to obtain is completely the same.Fig. 4 (g) carries out the segmentation result figure to the Hall of Supreme Harmony external door beam point cloud after next step process again after utilizing method of the present invention, can find out, after utilizing method of the present invention, can obtain good applicability to the Data Segmentation of ancient building point cloud.
In a further embodiment, as shown in Figure 5, the present invention acquires the partial data of The Gate of Supreme Harmony, comprises the crossbeam of a pillar and associated.Fig. 5 (a) comprises the cloud data of this pillar and this crossbeam.Counting of this Points Sample is 5771.Fig. 5 (b) is Gaussian mapping figure corresponding with it.Fig. 5 (c) is the N of AQ-DBSCAN algorithm programming count
mcurve map and the σ parameter value (0.014331) estimated.As can be seen from figure also, AQ-DBSCAN algorithm provides the result that σ value meets artificial judgment.Fig. 5 (d) is the Clustering Effect figure in Gaussian sphere, and because cylinder and crossbeam have overlap in Gaussian sphere, therefore also being gathered is a class.Fig. 5 (e) is the Clustering Effect figure corresponded on space curved surface.Fig. 5 (f) is the cluster adopting existing DBSCAN algorithm to obtain in Gaussian sphere, and as can be seen from Gaussian sphere, the Clustering Effect utilizing AQ-DBSCAN algorithm and DBSCAN algorithm to obtain is completely the same.Fig. 5 (g) carries out the segmentation result figure to the some cloud of The Gate of Supreme Harmony partial data after next step process again after utilizing method of the present invention, can find out, after utilizing method of the present invention, good applicability can be obtained to the Data Segmentation of ancient building point cloud.
Relative to DBSCAN algorithm, the present invention utilizes AQ-DBSCAN to carry out the method for area extension, and under the prerequisite keeping algorithm effect consistent, cluster speed is greatly improved.
The algorithm of three embodiments is consuming time as shown in table 1 above:
Table 1 DBSCAN, FDBSCAN, AQ-DBSCAN clustering algorithm consuming time
In figure 6 and figure 7, take time as horizontal ordinate, count as total coordinate makes curve consuming time with what process, curve A is the consuming time of DBSCAN algorithm in figure 6, curve B is the consuming time of AQ-DBSCAN algorithm, and curve C is the consuming time of FDBSCAN algorithm in the figure 7, and curve D is the consuming time of AQ-DBSCAN algorithm.Shown in Fig. 6 and Fig. 7 and table 1, as can be seen from consuming time relatively in, the consuming time of AQ-DBSCAN algorithm is obviously less than DBSCAN and FDBSCAN algorithm.Count more, the acceleration of AQ-DBSCAN algorithm is more obvious.Such as 141999 points, DBSCAN is consuming time is 6 times of AQ-DBSCAN, is 65 times for 642984 points.
Due to the singularity of ancient building point cloud, the dividing method of general neighborhood is made often to have larger unworthiness when applying.For these problems, a kind of clustering method being applicable to ancient building point cloud that the present invention proposes, is mapped to cloud data after Gaussian sphere is converted into 2-D data, carries out cluster to mapping point collection, filter out core point set, for the feature extraction in later stage and segmentation lay the first stone.
The present invention is the segmentation of all heavy constructions, useful exploration has been done in storage, expression, has important theory value and social effect.
Although embodiment of the present invention are open as above, but it is not restricted to listed in instructions and embodiment utilization, it can be applied to various applicable the field of the invention completely, for those skilled in the art, can easily realize other amendment, therefore do not deviating under the universal that claim and equivalency range limit, the present invention is not limited to specific details and illustrates here and the legend described.
Claims (7)
1. a clustering method for buildings Point Cloud of Laser Scanner, is characterized in that, comprising:
Step one, each point in the three dimensional point cloud obtained after carrying out laser scanning to buildings is converted into 2-D data, obtains the set X of the 2-D data of some cloud;
Step 2, specified density threshold value is minimum comprises the MinPts that counts, for any one point in set X, calculate and the nearest minimum maximum distance comprising MinPts object of counting of this point, and statistics to gather in X the maximal value of maximum distance a little and minimum value;
Step 3, the maximal value of maximum distance and the difference m of minimum value are divided into n equal portions, to produce the point of the minimum value of maximum distance in a cloud for the center of circle, in units of the spacing m/n of equal portions, increment value is radius step by step, makes n circle, calculates the quantity of point in each circle;
Step 4, using difference m as horizontal ordinate, interval as the incremental units of horizontal ordinate using the spacing m/n of equal portions, with the quantity put in each circle for ordinate, draw coordinate diagram, and the point on Fitting Coordinate System figure, form curve plotting;
The point of inflexion on a curve that step 5, searching are drawn, using the value of the numerical value of horizontal ordinate corresponding for this flex point as radius of neighbourhood σ;
Step 6, with density threshold MinPts and radius of neighbourhood σ for condition sets up AQ-DBSCAN algorithm, and with AQ-DBSCAN algorithm to set X in point carry out cluster, belong to the cluster analysis at which position in this buildings with the point obtained in described some cloud.
2. the clustering method of buildings Point Cloud of Laser Scanner as claimed in claim 1, it is characterized in that, described step 6 comprises:
6.1) to produce the point of the minimum value of maximum distance in a cloud for the center of circle, with radius of neighbourhood σ for radius, draw first order circle, if the number of the point in this first order circle is less than MinPts, then eliminate this first order circle, if be greater than MinPts; Then continue 6.2)
6.2) with c σ wherein 0 < c < 1 for radius, draw first order circle, if the first order is round and the number of point in annular region between the sub-circle of the first order is less than MinPts, then choose points all in annular region as the center of circle, the second level; If the number of the point in annular region is more than or equal to MinPts, then choose MinPts point as the center of circle, the second level;
6.3) with the selected center of circle, the second level for the center of circle, with radius of neighbourhood σ for radius, draw at least one second level circle, if the number of point in the circle of any one second level is less than MinPts, then eliminate this second level circle, if be greater than MinPts; Then repeated execution of steps 6.2) and 6.3), draw circle step by step, until the circle drawn all is eliminated;
6.4) being gathered by the point in the circle of all ranks is a class, and removes in described some cloud, and to the operation of the some repeated execution of steps two in remaining some cloud to step 6.
3. the clustering method of buildings Point Cloud of Laser Scanner as claimed in claim 2, is characterized in that, described step 6.2) in, choosing MinPts the method put as the center of circle, the second level is:
In this annular region, first choose from first order distance of center circle from farthest o'clock as first o'clock, then this distance of first of selected distance point is farthest as second point, then the distance sum of selected distance this first and second point point is farthest as thirdly, according to this rule until select MinPts point.
4. the clustering method of the buildings Point Cloud of Laser Scanner as described in claim 1 or 3, is characterized in that, described MinPts is 4.
5. the clustering method of buildings Point Cloud of Laser Scanner as claimed in claim 4, it is characterized in that, described c is 3/4.
6. the clustering method of buildings Point Cloud of Laser Scanner as claimed in claim 1, is characterized in that, described flex point is the maximum point of slope variation.
7. as claimed in claim 1 based on the clustering method of the some cloud of AQ-DBSCAN algorithm, it is characterized in that, in described step one, each point in the three dimensional point cloud obtained after carrying out laser scanning to buildings is mapped to Gaussian sphere thus is converted into 2-D data.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410539285.6A CN104573705B (en) | 2014-10-13 | 2014-10-13 | A kind of clustering method of building Point Cloud of Laser Scanner |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410539285.6A CN104573705B (en) | 2014-10-13 | 2014-10-13 | A kind of clustering method of building Point Cloud of Laser Scanner |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104573705A true CN104573705A (en) | 2015-04-29 |
| CN104573705B CN104573705B (en) | 2017-10-13 |
Family
ID=53089726
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410539285.6A Expired - Fee Related CN104573705B (en) | 2014-10-13 | 2014-10-13 | A kind of clustering method of building Point Cloud of Laser Scanner |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104573705B (en) |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780509A (en) * | 2016-12-01 | 2017-05-31 | 山东交通学院 | Merge the building object point cloud layer time cluster segmentation method of multidimensional characteristic |
| CN107392209A (en) * | 2017-07-11 | 2017-11-24 | 中国地质大学（武汉） | A kind of method, module, device and storage device for extracting line segment |
| CN108871268A (en) * | 2018-07-13 | 2018-11-23 | 湖南联智桥隧技术有限公司 | A kind of Tunnel Overbreak & Underbreak numerical computation method based on laser point cloud |
| CN108957476A (en) * | 2018-09-26 | 2018-12-07 | 中国计量大学 | A kind of Diameter measure device and tree-walk measuring method based on unmanned plane |
| CN109144072A (en) * | 2018-09-30 | 2019-01-04 | 亿嘉和科技股份有限公司 | A kind of intelligent robot barrier-avoiding method based on three-dimensional laser |
| CN109344812A (en) * | 2018-11-27 | 2019-02-15 | 武汉大学 | A kind of improved single photon point cloud data denoising method based on cluster |
| CN109523554A (en) * | 2018-11-16 | 2019-03-26 | 北京建筑大学 | A kind of ancient building point cloud automatic division method based on the wooden component |
| CN109948635A (en) * | 2017-12-21 | 2019-06-28 | 北京万集科技股份有限公司 | A kind of target identification method and device based on laser scanning |
| CN110096566A (en) * | 2019-06-18 | 2019-08-06 | 武汉中海庭数据技术有限公司 | A kind of determination method and system for the clustering parameter generating map lane line |
| CN110110802A (en) * | 2019-05-14 | 2019-08-09 | 南京林业大学 | Airborne laser point cloud classification method based on high-order condition random field |
| CN110246092A (en) * | 2019-05-02 | 2019-09-17 | 江西理工大学 | A kind of three-dimensional laser point cloud denoising method for taking neighborhood point mean distance and slope into account |
| CN111046330A (en) * | 2018-10-12 | 2020-04-21 | 中国电子科技集团公司第三十八研究所 | Method for generating effective points based on variable value space |
| CN111369609A (en) * | 2020-03-04 | 2020-07-03 | 山东交通学院 | Point cloud curved surface feature constraint-based building local deformation analysis method |
| CN111583257A (en) * | 2020-05-28 | 2020-08-25 | 中国铁道科学研究院集团有限公司 | Railway clearance foreign matter intrusion detection method, device and system |
| CN112070769A (en) * | 2020-09-18 | 2020-12-11 | 福州大学 | Layered point cloud segmentation method based on DBSCAN |
| CN112462347A (en) * | 2020-12-28 | 2021-03-09 | 长沙理工大学 | Laser radar point cloud rapid classification filtering algorithm based on density clustering |
| 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 |
| CN113076773A (en) * | 2020-01-03 | 2021-07-06 | 广州汽车集团股份有限公司 | Target detection clustering method, system, computer device and readable storage medium |
| CN115049028A (en) * | 2022-08-17 | 2022-09-13 | 中建五局第三建设有限公司 | Construction area partitioning method, system, terminal and medium based on unsupervised learning |
| CN116910888A (en) * | 2023-09-08 | 2023-10-20 | 临沂大学 | Method and system for generating BIM model component of assembled building |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101726255A (en) * | 2008-10-24 | 2010-06-09 | 中国科学院光电研究院 | Method for extracting interesting buildings from three-dimensional laser point cloud data |
| CN103614598A (en) * | 2013-12-20 | 2014-03-05 | 西南铝业（集团）有限责任公司 | Production method of aluminum alloy for rail-vehicle undercarriage plate |
| US20140098094A1 (en) * | 2012-10-05 | 2014-04-10 | Ulrich Neumann | Three-dimensional point processing and model generation |
-
2014
- 2014-10-13 CN CN201410539285.6A patent/CN104573705B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101726255A (en) * | 2008-10-24 | 2010-06-09 | 中国科学院光电研究院 | Method for extracting interesting buildings from three-dimensional laser point cloud data |
| US20140098094A1 (en) * | 2012-10-05 | 2014-04-10 | Ulrich Neumann | Three-dimensional point processing and model generation |
| CN103614598A (en) * | 2013-12-20 | 2014-03-05 | 西南铝业（集团）有限责任公司 | Production method of aluminum alloy for rail-vehicle undercarriage plate |
Non-Patent Citations (1)
| Title |
|---|
| 赵江洪等: "基于高斯映射的古建筑物点云分割", 《国际会议》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780509A (en) * | 2016-12-01 | 2017-05-31 | 山东交通学院 | Merge the building object point cloud layer time cluster segmentation method of multidimensional characteristic |
| CN107392209A (en) * | 2017-07-11 | 2017-11-24 | 中国地质大学（武汉） | A kind of method, module, device and storage device for extracting line segment |
| CN109948635A (en) * | 2017-12-21 | 2019-06-28 | 北京万集科技股份有限公司 | A kind of target identification method and device based on laser scanning |
| CN109948635B (en) * | 2017-12-21 | 2021-04-27 | 北京万集科技股份有限公司 | Target identification method and device based on laser scanning |
| CN108871268A (en) * | 2018-07-13 | 2018-11-23 | 湖南联智桥隧技术有限公司 | A kind of Tunnel Overbreak & Underbreak numerical computation method based on laser point cloud |
| CN108871268B (en) * | 2018-07-13 | 2021-02-02 | 湖南联智科技股份有限公司 | Tunnel under-excavation numerical calculation method based on laser point cloud |
| CN108957476A (en) * | 2018-09-26 | 2018-12-07 | 中国计量大学 | A kind of Diameter measure device and tree-walk measuring method based on unmanned plane |
| CN109144072A (en) * | 2018-09-30 | 2019-01-04 | 亿嘉和科技股份有限公司 | A kind of intelligent robot barrier-avoiding method based on three-dimensional laser |
| CN111046330A (en) * | 2018-10-12 | 2020-04-21 | 中国电子科技集团公司第三十八研究所 | Method for generating effective points based on variable value space |
| CN111046330B (en) * | 2018-10-12 | 2022-08-16 | 中国电子科技集团公司第三十八研究所 | Method for generating effective points based on variable value space |
| CN109523554A (en) * | 2018-11-16 | 2019-03-26 | 北京建筑大学 | A kind of ancient building point cloud automatic division method based on the wooden component |
| CN109344812A (en) * | 2018-11-27 | 2019-02-15 | 武汉大学 | A kind of improved single photon point cloud data denoising method based on cluster |
| CN109344812B (en) * | 2018-11-27 | 2021-06-04 | 武汉大学 | Improved cluster-based single photon point cloud data denoising method |
| CN110246092A (en) * | 2019-05-02 | 2019-09-17 | 江西理工大学 | A kind of three-dimensional laser point cloud denoising method for taking neighborhood point mean distance and slope into account |
| CN110246092B (en) * | 2019-05-02 | 2022-09-02 | 江西理工大学 | Three-dimensional laser point cloud denoising method considering neighborhood point mean distance and slope |
| CN110110802A (en) * | 2019-05-14 | 2019-08-09 | 南京林业大学 | Airborne laser point cloud classification method based on high-order condition random field |
| CN110096566A (en) * | 2019-06-18 | 2019-08-06 | 武汉中海庭数据技术有限公司 | A kind of determination method and system for the clustering parameter generating map lane line |
| 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 |
| CN113076773A (en) * | 2020-01-03 | 2021-07-06 | 广州汽车集团股份有限公司 | Target detection clustering method, system, computer device and readable storage medium |
| CN111369609A (en) * | 2020-03-04 | 2020-07-03 | 山东交通学院 | Point cloud curved surface feature constraint-based building local deformation analysis method |
| CN111369609B (en) * | 2020-03-04 | 2023-06-30 | 山东交通学院 | Building local deformation analysis method based on point cloud curved surface feature constraint |
| CN111583257A (en) * | 2020-05-28 | 2020-08-25 | 中国铁道科学研究院集团有限公司 | Railway clearance foreign matter intrusion detection method, device and system |
| CN112070769A (en) * | 2020-09-18 | 2020-12-11 | 福州大学 | Layered point cloud segmentation method based on DBSCAN |
| CN112070769B (en) * | 2020-09-18 | 2022-06-03 | 福州大学 | Layered point cloud segmentation method based on DBSCAN |
| CN112462347A (en) * | 2020-12-28 | 2021-03-09 | 长沙理工大学 | Laser radar point cloud rapid classification filtering algorithm based on density clustering |
| CN112462347B (en) * | 2020-12-28 | 2023-08-15 | 长沙理工大学 | Laser radar point cloud rapid classification filtering algorithm based on density clustering |
| CN115049028A (en) * | 2022-08-17 | 2022-09-13 | 中建五局第三建设有限公司 | Construction area partitioning method, system, terminal and medium based on unsupervised learning |
| CN115049028B (en) * | 2022-08-17 | 2022-12-13 | 中建五局第三建设有限公司 | Construction area partitioning method, system, terminal and medium based on unsupervised learning |
| CN116910888A (en) * | 2023-09-08 | 2023-10-20 | 临沂大学 | Method and system for generating BIM model component of assembled building |
| CN116910888B (en) * | 2023-09-08 | 2023-11-24 | 临沂大学 | Method and system for generating BIM model component of assembled building |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104573705B (en) | 2017-10-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN104573705A (en) | Clustering method for building laser scan point cloud data | |
| CN104392486B (en) | One kind point cloud scene reconstruction method | |
| Shi et al. | Adaptive simplification of point cloud using k-means clustering | |
| CN106157309B (en) | A kind of airborne LiDAR ground point cloud filtering method based on virtual seed point | |
| CN102890828B (en) | Point cloud data compacting method based on normal included angle | |
| US9507052B2 (en) | Automatic recognition method of continental slope foot point based on terrain grid | |
| CN105513127A (en) | Rod-shaped object regular three-dimensional modeling method and rod-shaped object regular three-dimensional modeling system based on density peak clustering | |
| CN106970375A (en) | A kind of method that building information is automatically extracted in airborne laser radar point cloud | |
| CN102136155A (en) | Object elevation vectorization method and system based on three dimensional laser scanning | |
| CN105719249A (en) | Three-dimensional grid-based airborne LiDAR point cloud denoising method | |
| CN106874580A (en) | A kind of bend pipe model reconstruction method based on cloud data | |
| CN105740798A (en) | Structure analysis based identification method for object in point cloud scene | |
| CN104766090B (en) | A kind of Coherent Noise in GPR Record method for visualizing based on BEMD and SOFM | |
| CN110147775A (en) | Utilize refinement method of the space separation method from data reduction indoor navigation element | |
| He et al. | A Method of Identifying Thunderstorm Clouds in Satellite Cloud Image Based on Clustering. | |
| Wang et al. | Fragmentation calculation method for blast muck piles in open-pit copper mines based on three-dimensional laser point cloud data | |
| CN102073867A (en) | Sorting method and device for remote sensing images | |
| Deng et al. | A novel fast classification filtering algorithm for LiDAR point clouds based on small grid density clustering | |
| CN103901467A (en) | Method for tracking positions of three-dimensional seismic data | |
| CN117370846B (en) | Hydraulic fracturing three-dimensional fracture network construction method and system based on microseism event | |
| CN104951752A (en) | Method for extracting houses from airborne laser point cloud data | |
| Zhang et al. | Object-based classification of urban airborne LiDAR point clouds with multiple echoes using SVM | |
| Wang et al. | Point cloud classification and accuracy analysis based on feature fusion | |
| Deng et al. | General multidimensional cloud model and its application on spatial clustering in Zhanjiang, Guangdong | |
| Zhao et al. | An automatic density clustering segmentation method for laser scanning point cloud data of buildings |
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 |
Granted publication date: 20171013 Termination date: 20211013 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |