CN105139379B - Based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface - Google Patents
Based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface Download PDFInfo
- Publication number
- CN105139379B CN105139379B CN201510465060.5A CN201510465060A CN105139379B CN 105139379 B CN105139379 B CN 105139379B CN 201510465060 A CN201510465060 A CN 201510465060A CN 105139379 B CN105139379 B CN 105139379B
- Authority
- CN
- China
- Prior art keywords
- top surface
- point
- initial
- sampled point
- accurate
- 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.)
- Active
Links
Classifications
-
- 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
- G06T2207/10032—Satellite or aerial image; Remote sensing
- G06T2207/10044—Radar image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2210/00—Indexing scheme for image generation or computer graphics
- G06T2210/04—Architectural design, interior design
Abstract
The invention discloses one kind to be based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface.Method is first classified to building top surface, is divided into " big top surface " and " small top surface " according to top surface area size, and top surface is divided into the different level of thickness according to corner dimension between top surface.On this basis, method using " from big to small ", " from coarse to fine ", the principle of " classification processing ", from LiDAR point cloud it is progressive extract building top surface.The region growing methods based on normal direction and the region growing methods based on distance are combined first, and big top surface is partitioned into from a cloud；Then cluster sub-clustering is carried out to left point, and small top surface is partitioned into from every cluster by random sampling coherence method.And by improving constantly angle Rule of judgment between top surface, progressively it is partitioned into the more tiny top surface of angle.Finally, automatic, the accurate segmentation of various building top surfaces is realized, so as to be laid the foundation for the automation modeling of three-dimensional building thing.
Description
Technical field
The present invention relates to the automatic Reconstruction technical field of earth's surface three-dimensional building thing model, and in particular to one kind is based on airborne
The progressive extracting method of Lidar point cloud building top surfaces, this method belongs to LiDAR point cloud data processing field, more particularly to is based on
The extraction of building top surface and three-dimensional model building reconstruct of Lidar point clouds.
Background technology
Realize that the automatic Reconstruction of earth's surface three-dimensional building thing model has great importance in many fields, such as 3-dimensional digital
Urban construction, city planning and administration, virtual tourism, or even risk assessment, contingency management etc..Using traditional surveying and mapping technology and
Method, although the geometric data of building can also be obtained and rebuild thirdly dimension module, because data acquisition speed is slow, modeling
The reasons such as process is cumbersome, cause traditional surveying and mapping technology to be difficult to the quick reconstruction demand for meeting extensive earth's surface scene.In the last few years,
As a kind of emerging active remote sensing surveying and mapping technology, i.e., airborne laser radar (Light Detection And Ranging,
LIDAR quick development) has been obtained and has been widely applied, it can quickly and accurately obtain the three-dimensional point of surface buildingses
Cloud data.It is to realize building model certainly that the Top-print information of building how is accurately partitioned into from these LIDAR cloud datas
The dynamic prerequisite and key link rebuild.Existing a variety of building top surface extracting methods based on different technologies thought at present.
The first kind is the building top surface extractive technique based on hierarchical cluster attribute.Generally, such technology is first according to sampled point
Neighborhood estimate the Differential Geometry attribute of each sampled point, such as normal direction, curvature etc.；Then by K-means clustering algorithms
(referring to：Sampath A,Shan J.Segmentation and reconstruction of polyhedral building
roofs from aerial lidar point clouds[J],IEEE Transactions on Geoscience and
Remote Sensing,2010,48(3):1554-1567) or K-plane clustering algorithms (such as Xu Lijun et al. invention " one
Building top surface point cloud data segmentation method of the kind based on K-plane algorithms ", number of patent application 201210478659.9) from
Segmentation obtains building top surface in Lidar point clouds.It is seen that the segmentation result of this kind of method is directly by Differential Geometry attribute
Calculate the influence of accuracy.However, in practice, the Differential Geometry attribute for accurately calculating each sampled point is the difficult thing of part
Feelings.When neighborhood choice is smaller, sampled point number is less in field, and the result of calculation of Differential Geometry attribute is influenceed by measurement error
It is larger；On the contrary, when neighborhood choice is larger, sampled point number is more in field, and it is real that Differential Geometry attribute is then difficult to reflection
Local detail feature.In addition, it is also one complicated that how such method, which when carrying out K-means clusters, sets initial K values,
Problem.
Second class is the building top surface extractive technique based on algorithm of region growing.Such technology is needed also exist for first according to neck
Domain estimates the Differential Geometry attributes such as the normal direction of each sampled point, curvature (or gradient, depth displacement, etc.)；Then, curvature is selected
Minimum sampled point is used as initial seed point, is made a living long constraints with the change size of normal direction and curvature, by outwards continuous
Growth by the basically identical sampled point of normal orientation be merged together and form a building top surface (referring to：Yu Haiyang, Yu Peng
Build, thank autumn equality, on-board LiDAR data building top surface point cloud segmentation technique study [J] mapping circulars, 2014 (6):20-
23).Such technology exist with the first kind technology identical deficiency, i.e., building roof segmentation accuracy by related differential category
Mutually calculate directly affecting for accuracy.Especially, this kind of method is difficult to the sampled point of accurate Ground Split top surface intersection, and face
The less top surface point of product.
3rd class is the building top surface extractive technique based on Model Matching.In this kind of method, most typically take out at random
Consistent (RANSAC) method of sample (referring to：Tarsha-Kurdi F,Landes T,Grussenmeyer P.Extended
RANSAC algorithm for automatic detection of building roof planes from LiDAR
data[J].The photogrammetric journal of Finland,2008,21(1):97-109).For comprising more
Individual top surface, or even the building Lidar cloud datas comprising noise point, this method by iteration judge can estimate one
The plane parameter model of maximum probability, and regard the sampled point for belonging to the model as a top surface split.This method
Deficiency is：First, the result of top surface segmentation is relevant with the order split, for baroque building, it is understood that there may be point
Error situation is cut, such as an actual top surface is divided into multiple top surfaces, multiple actual top surfaces are divided into top surface etc.；Its
It is secondary, when in Lidar point clouds include sampled point quantity it is more when, efficiency of algorithm is slow.
In fact, the top surface structure of most of surface buildingses is complicated and various, and the point cloud number that Lidar measurements obtain
According to it is at random, without topology, measurement error and noise even be present.Just because of this, adopts the same method, disposably segmentation is built
Thing top surface is difficult to obtain good effect.
The content of the invention
It can not only accurately extract that area is larger, boundary characteristic is brighter the technical problem to be solved in the present invention is to provide one kind
Aobvious building top surface, and can effectively identify and extract area is smaller, the less trickle building top surface of angle based on
The progressive extracting method of building top surface of classified and layered airborne Lidar cloud datas.
In order to solve the above technical problems, the present invention adopts the following technical scheme that：One kind is based on classified and layered airborne Lidar
The point progressive extracting method of cloud building top surface, it is characterised in that：Carry out according to the following steps,
1) classification layering, is carried out to building top surface, is on the one hand divided into " big top surface " and " small top according to top surface area size
The class of face " two, top surface is on the other hand divided into the different level of thickness according to corner dimension between top surface；
2), according to progressive extraction principle from big to small, from coarse to fine, first extract big top surface and extract small top surface again, first carry
The thick top surface that angle is larger is taken, then extracts the less trickle top surface of angle；Meanwhile different carry is taken different types of top surface
Take method；
3) after, often extracting a kind of top surface, the sampling extracted on top surface is removed from original building cloud data
Point, then other top surfaces are extracted from remaining sampled point；
4), parameter initialization, input point cloud data space be evenly distributed property coefficient, building top surface roughness value and
Cloud data sampling density, it is automatic to calculate the simultaneously relevant parameter initial value needed for method to set up；
5) the k- neighborhoods of sampled point, are searched, and estimate its normal direction and curvature：To any in building original point cloud data
Sampled point, searched by K-D data tree structures with it apart from k closest sampled point, and form its k- neighborhood；To k- neighborhoods
Sampled point set carry out principal component analysis, estimate the curvature and normal direction of the sampled point；
6), primary segmentation building original point cloud, and generate initial top surface set：Normal direction and curvature are set for constraint bar
Part, it is initial seed point to select the minimum sampled point of also undivided in original point cloud data and curvature, by region growing algorithm
By original building point cloud segmentation into some initial top surface set；
7) the larger initial big top surface of area, is selected, evaluates its coplanarity, and estimate its plane equation model：From initial
Chosen in top surface set comprising the more initial big top surface of sampled point, and take out its sampled point set；The initial top surface is adopted
Sampling point set carries out principal component analysis, calculates minimal eigenvalue and its corresponding characteristic vector；Evaluated according to minimal eigenvalue
The coplanarity of the initial big top surface；Characteristic vector and the geometric center point of sampled point set are drawn according to corresponding to minimal eigenvalue
The plane equation model of the initial big top surface；
8) the good initial big top surface of coplanarity, is selected, and extracts its accurate top surface：Select the first of coplanar sexual satisfaction requirement
Begin big top surface, takes out its plane equation model；Using the distance of sampled point to plane equation as constraints, with the initial big top surface
Existing sampled point be initial seed point, by the region growing algorithm based on distance, extract smart corresponding to the initial top surface
True top surface；
9) coplanarity and connectivity of accurate top surface, are detected, merges the accurate top surface for belonging to same actual top surface：To newly it carry
Compared with the accurate top surface taken out is carried out one by one with the accurate top surface extracted, if the normal direction phase of new top surface and some existing top surface
Closely and sampled point occurs simultaneously not for sky, then it is assumed that new top surface belongs to the same actual top surface of building with the existing top surface, and will be new
The sampled point of top surface is merged into the sampled point set of the existing top surface；Otherwise, new top surface is added in accurate top surface set；
10) coplanarity Rule of judgment, the smaller accurate top surface of iterative extraction angle, are improved：By circulate perform step 6),
7), 8) to this generation initial roof handling after the completion of, if exist coplanarity be unsatisfactory for require initial big top surface, improve
The Rule of judgment of coplanar point normal direction and curvature, jump to step 5) and primary segmentation is carried out to remaining sampled point again, and according to step
It is rapid 6), 7), 8) to continue to extract the smaller accurate top surface of angle, the initial big top required until no longer occurring coplanarity to be unsatisfactory for
Face；
11) cluster analysis, is carried out to remaining sampled point, is divided into more clusters：Removed from building original point cloud
It is extracted go out accurate big top surface on point；Using the Euclidean distance of point-to-point transmission as foundation, cluster analysis is carried out to remaining sampled point,
And it is divided into multiple clusters of diverse location；Every cluster is regarded as and belonged on a small top surface or multiple adjacent small top surfaces
Sampled point set；
12), iteration performs RANSAC algorithm, extracts its accurate small top surface included successively from every cluster：
Any cluster sampling point set is closed, iteration performs RANSAC algorithm, and extracts wherein include successively from big to small
Accurate small top surface.
It is further, as follows according to the specific extraction of preceding method and calculating process,
1) Lidar cloud data correlation attribute values, are inputted, it is automatic to calculate simultaneously arrange parameter initial value：
The correlation attribute value of Lidar cloud datas is inputted, including:Sampled point uniform spatial distribution property coefficient k1, building
Top surface roughness value k2And cloud data sampling density density；According to formulaAutomatically calculate and initialize k- Size of Neighborhood k value；According to formula kd=k2*(1/
(2*density1/2)) calculate coplanar point distance threshold kd；Size top surface judgment threshold k is set according to k valuess=k；Automatic is common
Millet cake normal direction threshold value knWith coplanar point curvature threshold kcOne larger initial value is set；
2) sampled point k- neighborhoods, are searched, and estimate its normal direction and curvature：
To any sampled point p in building original point cloud Pi(x, y, z), by K-D data tree structures search with its away from
From k closest sampled point, and form its k- neighborhood Nk(pi)；To Nk(pi) principal component analysis is carried out, characteristic value is calculated
λ0≥λ1≥λ2And characteristic vectorTo minimal characteristic λ2Corresponding characteristic vectorCarry out it is unitization, and by it
Approximate representation sampled point piThe normal direction at place；Utilize formula c=λ2/(λ0+λ1+λ2) calculate piThe approximate curvature c at place；
3), primary segmentation building original point cloud, and generate initial top surface set：
Iteration performs region growing algorithm, is partitioned into multiple initial top surfaces successively from primitive architecture object point cloud P, and form
Initial top surface set R0；Split initial top surface r performing region growing algorithm every timeiWhen, first from P select one it is also undivided,
And the sampled point of curvature minimum is initial seed point, and it is added in set in S；Then, to each seed point s in Si, look into
Its k- neighborhood is looked for, and judges any point p in its k- neighborhoodiWith siCoplanar implementations；If piAnd siNormal direction between angle it is small
In coplanar point normal direction threshold value kn, then by piIt is added to initial top surface riSampled point set P (ri) in；Meanwhile if piThe song of point
Rate is also less than coplanar point curvature threshold kc, then by piPoint regards a new seed point as, and is added to S；In this way constantly
Outwards increase, the point until no longer producing new seed, and generate an initial top surface ri；Iteration performs above-mentioned zone and increases calculation
Method, until there is no undivided cloud in P, then obtain initial top surface set R0；
4) the larger initial big top surface of area, is selected, judges its coplanarity, and estimate its plane equation model：
Judge R0In each initial top surface riIf P (ri) in sampled point quantity be more than ks, then it is assumed that riIt is larger for area
Initial top surface, referred to as initial big top surface；To each initial big top surface ri, take its sampled point set P (ri), and it is led
Constituent analysis, obtain minimal eigenvalue λ2And λ2Corresponding characteristic vectorAccording to λ2Value judge riIt is coplanar
Property, ifThen think riCoplanar sexual satisfaction requirement, and with A (x-x0)+B(y-y0)+C(z-z0The approximate table in)=0
Show riPlane equation mi；Otherwise, then it is assumed that riCoplanarity be unsatisfactory for requiring, and to riIt wouldn't handle；
5) the initial big top surface of coplanar sexual satisfaction requirement, is selected, and extracts its accurate top surface：
To any initial top surface r of coplanar sexual satisfaction requirementi, take out its sampled point set P (ri) and plane equation model
mi；With P (ri) it is initial seed point, with sampled point to miEuclidean distance be constraints, extracted by region growing algorithm
riCorresponding accurate summit ri′；Specific steps include, first, by P (ri) in sampled point regard initial seed point as and be added to kind
Sub- point set S；Secondly, each seed point S in S is searched one by oneiK- neighborhoods Nk(si)；Then, N is judgedk(si) in it is any one
Point p, if p to miDistance be less than kd, then it is assumed that p is the point on current top surface and is added to accurate top surface sampled point set p
(ri'), at the same time, if the k- neighborhoods N of p pointsk(p) in, there is point more than half to miDistance again smaller than kd, then it is assumed that p
For a new seed point, and it is added in S；Above-mentioned steps are performed, are constantly outwards increased, until no longer producing new kind
It is sub-, then stop increasing and by P (ri') in sampled point regard an accurate top surface r asi' on point；
6) coplanarity and connectivity of accurate top surface, are detected, merges the accurate top surface for belonging to same actual top surface：
To above-mentioned steps 5) the new accurate top surface r that extractsi', by itself and RaIn extracted each accurate top surface enter
Row compares, judge its whether with some existing accurate top surface rj' coplanar and connected；If so, then illustrate ri' and rj' belong to building
Same actual top surface, an accurate top surface need to be merged into；Otherwise, by ri' as a single new accurate top surface, add
It is added to RaIn；Here, judge ri' and rj' whether it is coplanar and connected when, the expression formula that uses for：And acos (ni·nj) ＜ ε, wherein rj′∈Ra、niAnd njFor ri' and rj' corresponding flat
The normal direction of equation model；
7) coplanarity Rule of judgment, the smaller accurate top surface of iterative extraction angle, are improved：
By circulate perform above-mentioned steps 4), 5), 6) to R0In each initial roof handling after the completion of, if existing coplanar
Property be unsatisfactory for desired initial big top surface, then improve the Rule of judgment of coplanar point normal direction and curvature, i.e., by knAnd kcValue halve；
Then, jump to step 3) and primary segmentation is carried out to remaining sampled point again, and continue to extract according to step 4), 5), 6) again
The smaller accurate top surface of angle, the initial big top surface required until no longer occurring coplanarity to be unsatisfactory for；
8) cluster analysis, is carried out to remaining sampled point, is divided into more clusters：
First R is removed from building original point cloud PaIn sampled point on each accurate top surface, then, with point-to-point transmission
Euclidean distance is foundation, carries out cluster analysis to remaining sampled point, and sampled point is divided into more cluster C=of diverse location
{ci, and will be per cluster ciRegard the sampled point set belonged on a small top surface or multiple adjacent small top surfaces as；
9), iteration performs RANSAC algorithm, extracts its accurate small top surface included successively from every cluster：
C is closed to any cluster sampling point seti, RANSAC algorithm is performed by iteration, carried successively from big to small
Take out the accurate small top surface wherein included；Specific steps include：First, the model of setting RANSAC algorithm proposition is
Space plane model, judge that a sampled point belongs to areal model internal point distance threshold d=kd；Secondly, by random sampling
Consistency algorithm is from ciMiddle extraction top surface r one smalli, including riPlane in terms of model miWith internal point set P (ri)；Then,
Judge riReasonability, if miC has been crossed in three dimensionsi, it is believed that riIt is unreasonable, then by RANSAC algorithm
Constraints d values halve, otherwise it is assumed that riRationally, then by riIt is added to RaIn and from ciMiddle removal P (ri)；Finally, c is judgediIn
Sampled point whether be less than 4, if it is not, then repeating above-mentioned steps continues to extract other accurate small top surfaces, if so, then stopping
Extract and by RaRegard final building top surface extraction result as.
Wherein, P={ pi(x, y, z) }, represent building crude sampling point set, wherein Pi(x, y, z) represents Lidar points
Sampled point in cloud；
Nk(pi), represent piThe k- neighborhoods of point, i.e. distance piThe set of k nearest sampled point；
R, represent top surface, riRepresent an initial top surface, ri' represent riCorresponding accurate top surface；
P (r) represents the sampling point set for belonging to top surface r, P (ri) represent to belong to initial top surface riPoint set, P (ri') represent category
In accurate top surface ri' point set；
R, represent top surface set, R0={ riRepresent initial top surface set, Ra={ ri' represent accurate top surface set；
M, represent plane equation model, miRepresent initial top surface riPlane equation model；
Building crude sampling point set P={ pi(x, y, z) }；
Lidar cloud data sampling density density, unit are individual/m2；
Sampled point uniform spatial distribution property coefficient k1, span be (0,1], the more uniform k of sampling point distributions1Value is got over
Greatly；
Building top surface roughness value k2, span be (0,1], construction ceiling is more smooth, k2Value is bigger；
Extract obtained accurate top surface set Ra={ ri′}。
The present invention is based on classified and layered airborne Lidar cloud datas, and the base of classification is being carried out to building top surface
On plinth, using " from big to small ", the approach principle of " from coarse to fine ", building is realized by way of classifying and handling and progressively refine
Automatic, the accurate segmentation of thing top surface, and then extract the accurate top surface of the building included in Lidar point clouds.
Brief description of the drawings
Fig. 1 is flow chart of the present invention；
Fig. 2 is the original point cloud data of the building arrived by airborne Lidar system acquisitions；
Fig. 3 is point clouds of the building original point cloud P after first time primary segmentation；
Fig. 4 is the accurate top surface point cloud obtained by extracting big top surface；
Fig. 5 is the smaller accurate top surface of extraction angle；
Fig. 6 is to extract the accurate top surface that big top surface and the smaller top surface of angle newly obtain initial top surface；
Fig. 7 is remaining small top surface point cloud cluster result；
Segmentation result final Fig. 8.
Embodiment
The Lidar cloud datas of real building thing selected by the present embodiment, the building altogether comprising A, B, C, D, E, F, G,
H, the top surface of 12 sizes such as I, J, L, Q not etc., and the angle between top surface also differs in size.For example, Q, L, I and J tetra-
Top surface area very little (sampled point number thereon is only 5,6,6,5), the angle very little between two top surfaces of D, C.Together
When, have above the building setting elongated chimney and two horizontal through electric wire, this can cause the building
Lidar cloud datas carry noise point.
The original Lidar cloud datas situation of building is as shown in Fig. 2 which show arrived by airborne Lidar system acquisitions
The building original point cloud data, wherein, the pitch black point in L regions is the point on the minimum top surface of the building；The depth in H regions
Stain is building upper horizontal electric wire and erects the point on chimney, belongs to peak；The secondary stain in middle M regions is the building
The high point at thing median rise position.The Points Sample density is every square metre of 4 sampled points.Meanwhile the cloud data sampled point
Spatial distribution is also more smooth than more uniform and building top surface.
Flow shown in reference picture 1, the specific steps of top surface extraction are carried out such as to above-mentioned building Lidar cloud datas
Under：
1st, Lidar cloud data characteristic values are inputted, it is automatic to calculate simultaneously arrange parameter initial value；
Set the sampled point uniform spatial distribution degree coefficient k of building Lidar cloud datas1=0.8, construction ceiling
Coefficient of roughness k2=0.6, according to formulaK- Size of Neighborhood k=9 is calculated；Set according to k values
Size top surface judgment threshold ks=9；According to sampling density (4/m2) and the construction ceiling coefficient of roughness (k2=0.6), according to public affairs
Formula kd=k2*(1/(2*density1/2)) calculate coplanar point distance threshold kd=0.15 meter；It is individual according to k- Size of Neighborhood k=9,
It is initialized as size top surface judgment threshold ks=9；Setting initialization coplanar point normal direction threshold value kn=8 °, coplanar point curvature threshold
kc=0.3.
2nd, the k- neighborhoods of sampled point are searched, and estimate its normal direction and curvature；
To any sampled point p in building original point cloud Pi(x, y, z), by the lookup of K-D data tree structures and its
Apart from k closest sampled point, and form its k- neighborhood Nk(pi)；Calculate Nk(pi) cluster center in three dimensions
PointConstruct covariance matrix C：Calculate the eigenvalue λ of Matrix C0≥λ1
≥λ2And its corresponding characteristic vectorWillCarry out it is unitization, and by its approximate representation sampled point piThe normal direction at place；
Utilize formula c=λ2/(λ0+λ1+λ2) calculate piThe approximate curvature c at place.
3rd, primary segmentation building original point cloud, and generate initial top surface set；
Iteration performs region growing algorithm, is partitioned into multiple initial top surfaces successively from the original point cloud P of the building, and
Form initial top surface set R0.Detailed step includes：
A. initial top surface r is definediSampled point setSeed point setInitial top surface set
B. the minimum sampled point of undivided in original point cloud P and curvature is found out, and is regarded as initial seed point and is added to
In set in S；
C. the seed point s in S is taken out successively from front to backi, and search its k- neighborhood sampled point P (si)；
D. P (s are judged successivelyi) in any point piIf piAnd siThe angle of normal direction be less than coplanar point normal direction threshold value
kn, by piIt is denoted as having split and is added to P (ri) in；Meanwhile if piThe curvature of point is less than coplanar point curvature threshold kc, by pi
It is added in S；
E. until not being further added by new seed point in S, by riRegard that an initial top surface is added to R as0In；
F. by P (ri) and S be re-set as sky, judge in original point cloud P whether to also have undivided sampled point, if so,
Then re-execute step b, c, d, e and extract next initial top surface, otherwise terminate to split and obtain initial top surface set R0。
Obtained result is as shown in figure 3, give the original point cloud P of building first time primary segmentation result.
Gray scale is identical in the figure and one initial top surface of continuous point expression, black point represent temporarily to fail the sampling correctly split
Point.As can be seen that 10 initial top surfaces have been obtained in this, and the actual top surface of this 10 initial top surfaces and building has three
Kind relation：(1) one-one relationship:The corresponding actual top surface of i.e. one initial top surface, such as the corresponding buildings of initial top surface F '
One actual top surface, but it is seen that initial top surface fails to include actual top edge sampled point；(2) many-one relationship:I.e. one
Individual initial top surface correspond to multiple actual top surfaces, as initial top surface C ' to should building two adjacent actual top surfaces；(3) it is multipair
One relation：I.e. multiple corresponding actual top surfaces of initial top surfaces, such as initial top surface A ' and A ' ' to should building same reality
Top surface.
4th, the larger initial big top surface of area is selected, evaluates its coplanarity, and estimates model in terms of its plane；
The initial top surface set R that judgment step 3 obtains0In each initial top surface riIf P (ri) it is more than ks, then it is assumed that ri
For the initial top surface that area is larger, referred to as initial big top surface.To each initial big top surface ri, calculate P (ri) geometric center pointAnd according to P (ri) andConstruct covariance matrixSolve Matrix C
Minimal eigenvalue λ2And its corresponding characteristic vectorAccording to minimal eigenvalue λ2Judge riCoplanarity, ifThen think riCoplanar sexual satisfaction requirement, and with a (x-x0)+b(y-y0)+c(z-z0Approximate representation r is carried out in)=0iPlane
Equation mi；Otherwise, then it is assumed that riCoplanarity be unsatisfactory for requiring, and to riIt wouldn't handle.In the initial top surface of the building, estimation
Obtain initial top surface C ' and be unsatisfactory for requiring initial top surface for coplanarity, it is seen that to contain two angles smaller by initial top surface C '
Actual top surface.
5th, the good initial big top surface of coplanarity is selected, and extracts its accurate top surface；
To any initial top surface r of coplanar sexual satisfaction requirementi, take its sampled point set P (ri) and plane equation model mi,
Accurate top surface r corresponding to it is extracted from building object point cloud P by region growing algorithmi' sampled point set P (ri'), specifically
Step is as follows：
A. accurate top surface r is definedi' sampled point setSeed point set
B. by P (ri) in sampled point regard initial seed point as and be added in set S；
C. the seed point s in S is taken out successively from front to backi, and search siK- neighborhood sampled points Nk(si)；
D. N is judgedk(si) in any point p, if p to miDistance be less than kdAnd p is not in current P (ri') in, then recognize
It is point on current top surface for p and is added to accurate top surface sampled point set P (ti'), at the same time, if the k- neighborhoods N of p pointsk
(p) in, there is point more than half to miDistance again smaller than kd, then it is assumed that p is new seed (seed) point, and is added
It is added in S；
E. until not being further added by new seed point in S, then stop increasing and by P (ri') regard accurate top surface r asi' adopt
Sampling point.
6th, the coplanarity and connectivity of accurate top surface are detected, merges the accurate top surface for belonging to same actual top surface；
The new accurate top surface r extracted to step 5i', by itself and RaIn extracted each accurate top surface compared
Compared with judging it, whether accurate top surface rj ' is coplanar and is connected with some.If so, then illustrate ri' and rj' belong to the same of building
One actual top surface, an accurate top surface need to be merged into；Otherwise, by ri' as a single new accurate top surface, it is added to
RaIn.Here, judge ri' and rj' whether it is coplanar and connected when, the expression formula that uses for：And
acos(ni·nj) ＜ ε, wherein rj′∈Ra、niAnd njFor ri' and rjThe normal direction of ' corresponding flat equation model.
It is as shown in Figure 4 that obtained accurate top surface is extracted afterwards by step 5,6.The accurate top surface for now extracting to obtain is with building
The actual top surface for building thing is one-one relationship, and the sampled point of accurate top surface and actual top surface is also basically identical.
7th, coplanarity Rule of judgment, the smaller accurate top surface of iterative extraction angle are improved；
In the building, because initial top surface C ' coplanarity is unsatisfactory for requiring, therefore algorithm is by current knAnd kcValue
After halving, primary segmentation is continued to remaining sampled point according to step 3, segmentation result is as shown in figure 5, this time by the building
The middle less two actual top surfaces of angle have been divided into 3 initial top surface C ', D ' and D ' '.On this basis, by step 4,5,
6 extractions newly obtain the accurate top surface of initial top surface, and final result of extracting is as shown in fig. 6, now will be big in the building
Top surface all extracts, and the pitch black point in figure top is the sampled point or noise point on small top surface.
8th, cluster analysis is carried out to remaining sampled point, is divided into more clusters；
Using the Euclidean distance of point-to-point transmission as foundation, cluster analysis is carried out to the remaining sampled point in P, sampled point is divided into
Multiple cluster C={ c of diverse locationi}.Cluster result is as shown in fig. 7, C1And C2Two cluster sampled points correspond to small in building respectively
Top surface.
9th, iteration performs RANSAC algorithm, extracts its accurate small top surface included successively from every cluster；
C is closed to any cluster sampling point seti, RANSAC algorithm (RANSAC algorithms) is performed by iteration, from
The small accurate small top surface for extracting wherein include successively is arrived greatly.Specific steps include：
A. the types of models for setting the estimation of RANSAC algorithms is space plane model, sets RANSAC algorithms to judge that one is adopted
Whether sampling point belongs to the distance threshold d=k of current planed；
B. by RANSAC algorithms from ciMiddle extraction top surface r one smalli, including riPlane in terms of model miAnd internal point
Set P (ri)；
C. r is judgediReasonability, if miC has been crossed in three dimensionsi, it is believed that riIt is unreasonable, then RANSAC is calculated
Method constraints d values halve, otherwise it is assumed that riRationally, then by riIt is added to RaIn and from ciMiddle removal P (ri)；
D. c is judgediIn sampled point whether be less than 4, if it is not, then repeat above-mentioned steps continue to extract it is other accurate
Small top surface, if so, then stopping extraction and by RaRegard final building top surface extraction result as.
By above-mentioned steps, building top surface extraction is final to be extracted that area is larger, border is significantly pushed up exactly
Face, and effectively extracted that area is smaller, top surface of obscurity boundary, while the interference of noise point is avoided, final segmentation knot
Fruit is as shown in Figure 8.
The present invention is described in detail above, described above, only the preferred embodiments of the invention, when can not
Limit the scope of the present invention, i.e., it is all to make equivalent changes and modifications according to the application scope, it all should still belong to covering scope of the present invention
It is interior.
Claims (2)
1. one kind is based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface, it is characterised in that：By following
Step is carried out,
1) classification layering, is carried out to building top surface, is on the one hand divided into " big top surface " and " small top surface " according to top surface area size
Two classes, top surface is on the other hand divided into the different level of thickness according to corner dimension between top surface；
2), according to progressive extraction principle from big to small, from coarse to fine, first extract big top surface and extract small top surface again, first extraction folder
The larger thick top surface in angle, then extract the less trickle top surface of angle；Meanwhile different extraction sides is taken to different types of top surface
Method；
3) after, often extracting a kind of top surface, the sampled point extracted on top surface is removed from original building cloud data, then
Other top surfaces are extracted from remaining sampled point；
4), parameter initialization, input point cloud data space be evenly distributed property coefficient, building top surface roughness value and point cloud
Data sampling density, it is automatic to calculate the simultaneously relevant parameter initial value needed for method to set up；
5) the k- neighborhoods of sampled point, are searched, and estimate its normal direction and curvature：To any sampling in building original point cloud data
Point, searched by K-D data tree structures with it apart from k closest sampled point, and form its k- neighborhood；K- neighborhoods are adopted
Sampling point set carries out principal component analysis, estimates the curvature and normal direction of the sampled point；
6), primary segmentation building original point cloud, and generate initial top surface set：It is constraints to set normal direction and curvature, choosing
It is initial seed point to select the minimum sampled point of also undivided in original point cloud data and curvature, by region growing algorithm by original
Establish and build object point cloud and be divided into some initial top surface set；
7) the larger initial big top surface of area, is selected, evaluates its coplanarity, and estimate its plane equation model：From initial top surface
Chosen in set comprising the more initial big top surface of sampled point, and take out its sampled point set；To the sampled point of the initial top surface
Set carries out principal component analysis, calculates minimal eigenvalue and its corresponding characteristic vector；It is first that this is evaluated according to minimal eigenvalue
Begin the coplanarity of big top surface；Characteristic vector and the geometric center point of sampled point set show that this is first according to corresponding to minimal eigenvalue
Begin the plane equation model of big top surface；
8) the good initial big top surface of coplanarity, is selected, and extracts its accurate top surface：Select the initial big of coplanar sexual satisfaction requirement
Top surface, take out its plane equation model；Using the distance of sampled point to plane equation as constraints, with the initial big top surface
It is initial seed point to have sampled point, by the region growing algorithm based on distance, extracts and is accurately pushed up corresponding to the initial top surface
Face；
9) coplanarity and connectivity of accurate top surface, are detected, merges the accurate top surface for belonging to same actual top surface：To newly it extract
Accurate top surface carried out one by one with the accurate top surface that has extracted compared with, if new top surface it is close with the normal direction that some has top surface,
And sampled point occurs simultaneously not for sky, then it is assumed that new top surface belongs to the same actual top surface of building with the existing top surface, and will newly push up
The sampled point in face is merged into the sampled point set of the existing top surface；Otherwise, new top surface is added in accurate top surface set；
10) coplanarity Rule of judgment, the smaller accurate top surface of iterative extraction angle, are improved：By circulate perform step 6), 7),
8) after the completion of to the initial roof handling of this generation, if the initial big top surface that coplanarity is unsatisfactory for requiring be present, improve altogether
The Rule of judgment of millet cake normal direction and curvature, jump to step 5) and primary segmentation is carried out to remaining sampled point again, and according to step
6), 7), 8) continue to extract the smaller accurate top surface of angle, the initial big top surface required until no longer occurring coplanarity to be unsatisfactory for；
11) cluster analysis, is carried out to remaining sampled point, is divided into more clusters：Remove and carried from building original point cloud
The point on accurate big top surface taken out；Using the Euclidean distance of point-to-point transmission as foundation, cluster analysis is carried out to remaining sampled point, and will
It is divided into multiple clusters of diverse location；Every cluster is regarded as to the sampling belonged on a small top surface or multiple adjacent small top surfaces
Point set；
12), iteration performs RANSAC algorithm, extracts its accurate small top surface included successively from every cluster：To appointing
Meaning cluster sampled point set, iteration performs RANSAC algorithm, and extracts the essence wherein included successively from big to small
True small top surface.
2. according to claim 1 be based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface, its
It is characterised by：Specific extraction and calculating process are as follows,
1) Lidar cloud data correlation attribute values, are inputted, it is automatic to calculate simultaneously arrange parameter initial value：
The correlation attribute value of Lidar cloud datas is inputted, including:Sampled point uniform spatial distribution property coefficient k1, building top surface it is thick
Roughness coefficient k2And cloud data sampling density density；According to formulaAutomatically
Calculate and initialize k- Size of Neighborhood k value；According to formula kd=k2*(1/(2*density1/2)) coplanar point is calculated apart from threshold
Value kd；Size top surface judgment threshold k is set according to k valuess=k；Automatic is coplanar point normal direction threshold value knWith coplanar point curvature threshold kc
One larger initial value is set；
2) sampled point k- neighborhoods, are searched, and estimate its normal direction and curvature：
To any sampled point p in building original point cloud Pi(x, y, z), searched by K-D data tree structures most adjacent with its distance
K near sampled point, and form its k- neighborhood Nk(pi)；To Nk(pi) principal component analysis is carried out, eigenvalue λ is calculated0≥λ1
≥λ2And characteristic vectorTo minimal characteristic λ2Corresponding characteristic vectorCarry out it is unitization, and by its approximate table
Show sampled point piThe normal direction at place；Utilize formula c=λ2/(λ0+λ1+λ2) calculate piThe approximate curvature c at place；
3), primary segmentation building original point cloud, and generate initial top surface set：
Iteration performs region growing algorithm, is partitioned into multiple initial top surfaces successively from primitive architecture object point cloud P, and forms initial
Top surface set R0；Split initial top surface r performing region growing algorithm every timeiWhen, first selection one is also undivided and bent from P
The minimum sampled point of rate is initial seed point, and is added in set in S；Then, to each seed point s in Si, search it
K- neighborhoods, and judge any point p in its k- neighborhoodiWith siCoplanar implementations；If piAnd siNormal direction between angle be less than altogether
Millet cake normal direction threshold value kn, then by piIt is added to initial top surface riSampled point set P (ri) in；Meanwhile if piThe curvature of point is also
Less than coplanar point curvature threshold kc, then by piPoint regards a new seed point as, and is added to S；It is constantly outside in this way
Increase, the point until no longer producing new seed, and generate an initial top surface ri；Iteration performs above-mentioned zone growth algorithm,
Until there is no undivided cloud in P, then initial top surface set R is obtained0；
4) the larger initial big top surface of area, is selected, judges its coplanarity, and estimate its plane equation model：
Judge R0In each initial top surface riIf P (ri) in sampled point quantity be more than ks, then it is assumed that riFor area it is larger just
Beginning top surface, referred to as initial big top surface；To each initial big top surface ri, take its sampled point set P (ri), and principal component is carried out to it
Analysis, obtains minimal eigenvalue λ2And λ2Corresponding characteristic vectorAccording to λ2Value judge riCoplanarity,
IfThen think riCoplanar sexual satisfaction requirement, and with A (x-x0)+B(y-y0)+C(z-z0Approximate representation r is carried out in)=0i
Plane equation mi；Otherwise, then it is assumed that riCoplanarity be unsatisfactory for requiring, and to riIt wouldn't handle；
5) the initial big top surface of coplanar sexual satisfaction requirement, is selected, and extracts its accurate top surface：
To any initial top surface r of coplanar sexual satisfaction requirementi, take out its sampled point set P (ri) and plane equation model mi；With P
(ri) it is initial seed point, with sampled point to miEuclidean distance be constraints, by region growing algorithm extract riIt is corresponding
Accurate summit r 'i；Specific steps include, first, by P (ri) in sampled point regard initial seed point as and be added to seed point set
Close S；Secondly, each seed point s in S is searched one by oneiK- neighborhoods Nk(si)；Then, N is judgedk(si) in any point p, such as
Fruit p to miDistance be less than kd, then it is assumed that p is the point on current top surface and is added to accurate top surface sampled point set P (ri'), with
This simultaneously, if the k- neighborhoods N of p pointsk(p) in, there is point more than half to miDistance again smaller than kd, then it is assumed that p is one new
Seed point, and be added in S；Above-mentioned steps are performed, are constantly outwards increased, until no longer producing new seed point, then
Stop increasing and by P (ri') in sampled point regard an accurate top surface r asi' on point；
6) coplanarity and connectivity of accurate top surface, are detected, merges the accurate top surface for belonging to same actual top surface：
To above-mentioned steps 5) the new accurate top surface r that extractsi', by itself and RaIn extracted each accurate top surface compared
Compared with, judge its whether with some existing accurate top surface rj' coplanar and connected；If so, then illustrate ri' and rj' belong to the same of building
One actual top surface, an accurate top surface need to be merged into；Otherwise, by ri' as a single new accurate top surface, it is added to
RaIn；Here, judge ri' and rj' whether it is coplanar and connected when, the expression formula that uses for：
And acos (ni·nj) ＜ ε, wherein rj′∈Ra、niAnd njFor ri' and rjThe normal direction of ' corresponding flat equation model；
7) coplanarity Rule of judgment, the smaller accurate top surface of iterative extraction angle, are improved：
By circulate perform above-mentioned steps 4), 5), 6) to R0In each initial roof handling after the completion of, if coplanarity be present not
Meet desired initial big top surface, then improve the Rule of judgment of coplanar point normal direction and curvature, i.e., by knAnd kcValue halve；Then,
Jump to step 3) and carry out primary segmentation to remaining sampled point again, and continue extraction angle more according to step 4), 5), 6) again
Small accurate top surface, the initial big top surface required until no longer occurring coplanarity to be unsatisfactory for；
8) cluster analysis, is carried out to remaining sampled point, is divided into more clusters：
First R is removed from building original point cloud PaIn sampled point on each accurate top surface, then, with the European of point-to-point transmission
Distance is foundation, carries out cluster analysis to remaining sampled point, and sampled point is divided into more cluster C={ c of diverse locationi, and
Will be per cluster ciRegard the sampled point set belonged on a small top surface or multiple adjacent small top surfaces as；
9), iteration performs RANSAC algorithm, extracts its accurate small top surface included successively from every cluster：
C is closed to any cluster sampling point seti, RANSAC algorithm is performed by iteration, extracts it successively from big to small
In the accurate small top surface that includes；Specific steps include：First, the model for setting RANSAC algorithm to propose is put down for space
Surface model, judge that a sampled point belongs to areal model internal point distance threshold d=kd；Secondly, by random sampling uniformity
Algorithm is from ciMiddle extraction top surface r one smalli, including riPlane in terms of model miWith internal point set P (ri)；Then, r is judgedi
Reasonability, if miC has been crossed in three dimensionsi, it is believed that riIt is unreasonable, then RANSAC algorithm is constrained into bar
Part d values halve, otherwise it is assumed that riRationally, then by riIt is added to RaIn and from ciMiddle removal P (ri)；Finally, c is judgediIn sampling
Whether point is less than 4, continues to extract other accurate small top surfaces if it is not, then repeating above-mentioned steps, if so, then stopping extraction simultaneously
By RaRegard final building top surface extraction result as.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510465060.5A CN105139379B (en) | 2015-07-30 | 2015-07-30 | Based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510465060.5A CN105139379B (en) | 2015-07-30 | 2015-07-30 | Based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105139379A CN105139379A (en) | 2015-12-09 |
| CN105139379B true CN105139379B (en) | 2018-01-12 |
Family
ID=54724712
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510465060.5A Active CN105139379B (en) | 2015-07-30 | 2015-07-30 | Based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105139379B (en) |
Families Citing this family (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127857B (en) * | 2016-06-24 | 2018-11-16 | 南京林业大学 | The on-board LiDAR data modeling method of integrated data driving and model-driven |
| CN107993242B (en) * | 2017-12-14 | 2022-06-03 | 天津大学 | Method for extracting boundary of missing area based on airborne LiDAR point cloud data |
| CN107977968B (en) * | 2017-12-22 | 2021-03-19 | 长江勘测规划设计研究有限责任公司 | Building layered detection method based on building shadow information mining |
| CN108572361A (en) * | 2018-04-03 | 2018-09-25 | 深圳飞马机器人科技有限公司 | Airborne laser radar system equipment integrates angle of setting calibration method and device |
| CN108898596B (en) * | 2018-06-28 | 2020-09-04 | 中国地震局地震预测研究所 | Method for detecting destruction state of building |
| CN109670557B (en) * | 2019-01-03 | 2023-07-21 | 中国矿业大学 | Automatic highway point cloud registration method based on rod-shaped structures |
| CN110095062B (en) * | 2019-04-17 | 2021-01-05 | 北京华捷艾米科技有限公司 | Object volume parameter measuring method, device and equipment |
| CN112285733B (en) * | 2020-10-21 | 2023-09-26 | 中核勘察设计研究有限公司 | Urban and rural planning verification mapping data processing method |
| CN112802093B (en) * | 2021-02-05 | 2023-09-12 | 梅卡曼德（北京）机器人科技有限公司 | Object grabbing method and device |
| CN113205529B (en) * | 2021-04-19 | 2022-04-29 | 武汉大学 | Method for segmenting top surface of building based on airborne LiDAR point cloud |
| CN113409332B (en) * | 2021-06-11 | 2022-05-27 | 电子科技大学 | Building plane segmentation method based on three-dimensional point cloud |
-
2015
- 2015-07-30 CN CN201510465060.5A patent/CN105139379B/en active Active
Non-Patent Citations (5)
| Title |
|---|
| A global optimization approach to roof segmentation from airborne lidar point clouds;Jixing Yan et al;《ISPRS Journal of Photogrammetry and Remote Sensing》;20140604;第94卷;全文 * |
| LIDAR点云数据的建筑物特征线提取;孟峰 等;《测绘科学》;20080930;第33卷(第5期);全文 * |
| 地面激光扫描数据在建筑物重建中的研究进展;王方建 等;《遥感信息》;20141231;第29卷(第6期);全文 * |
| 基于FCM的机载LIDAR数据的建筑物和植被分类方法;刘昌军 等;《中国水利水电科学研究院学报》;20130930;第11卷(第3期);全文 * |
| 机载LiDAR数据建筑物顶面点云分割方法研究;于海洋 等;《测绘通报》;20141231(第6期);全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105139379A (en) | 2015-12-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105139379B (en) | Based on the progressive extracting method of classified and layered airborne Lidar points cloud building top surface | |
| Xu et al. | Reconstruction of scaffolds from a photogrammetric point cloud of construction sites using a novel 3D local feature descriptor | |
| US20220398805A1 (en) | Large-scale environment-modeling with geometric optimization | |
| CN102938066B (en) | A kind of based on the polygonal method of multivariate data reconstruction buildings outline | |
| Rottensteiner et al. | A new method for building extraction in urban areas from high-resolution LIDAR data | |
| CN111815776A (en) | Three-dimensional building fine geometric reconstruction method integrating airborne and vehicle-mounted three-dimensional laser point clouds and streetscape images | |
| CN106023298B (en) | Point cloud Rigid Registration method based on local Poisson curve reestablishing | |
| CN111815757A (en) | Three-dimensional reconstruction method for large component based on image sequence | |
| CN108171780A (en) | A kind of method that indoor true three-dimension map is built based on laser radar | |
| US9942535B2 (en) | Method for 3D scene structure modeling and camera registration from single image | |
| Ramiya et al. | Segmentation based building detection approach from LiDAR point cloud | |
| Wang et al. | Modeling indoor spaces using decomposition and reconstruction of structural elements | |
| CN114332366A (en) | Digital city single house point cloud facade 3D feature extraction method | |
| Lee et al. | Automatic integration of facade textures into 3D building models with a projective geometry based line clustering | |
| CN102804231A (en) | Piecewise planar reconstruction of three-dimensional scenes | |
| Canaz Sevgen et al. | An improved RANSAC algorithm for extracting roof planes from airborne lidar data | |
| CN110222642A (en) | A kind of planar architectural component point cloud contour extraction method based on global figure cluster | |
| Lafarge et al. | Insertion of 3-D-primitives in mesh-based representations: Towards compact models preserving the details | |
| CN101901343A (en) | Remote sensing image road extracting method based on stereo constraint | |
| CN108830776A (en) | The visible entity watermark copyright anti-counterfeiting mark method of three-dimensional towards 3D printing model | |
| Shan et al. | Building extraction from LiDAR point clouds based on clustering techniques | |
| Song et al. | Extraction and reconstruction of curved surface buildings by contour clustering using airborne LiDAR data | |
| CN110807781A (en) | Point cloud simplification method capable of retaining details and boundary features | |
| CN103324916A (en) | Registration method for vehicle-mounted LiDAR data and aviation LiDAR data based on building outlines | |
| CN104751479A (en) | Building extraction method and device based on TIN data |
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 |