CN114742957B - Building facade extraction method based on point cloud data - Google Patents

Building facade extraction method based on point cloud data Download PDF

Info

Publication number
CN114742957B
CN114742957B CN202210659698.2A CN202210659698A CN114742957B CN 114742957 B CN114742957 B CN 114742957B CN 202210659698 A CN202210659698 A CN 202210659698A CN 114742957 B CN114742957 B CN 114742957B
Authority
CN
China
Prior art keywords
facade
point cloud
plane
potential
cloud data
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
Application number
CN202210659698.2A
Other languages
Chinese (zh)
Other versions
CN114742957A (en
Inventor
于冰
胡金龙
胡云亮
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Southwest Petroleum University
Original Assignee
Southwest Petroleum University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Southwest Petroleum University filed Critical Southwest Petroleum University
Priority to CN202210659698.2A priority Critical patent/CN114742957B/en
Publication of CN114742957A publication Critical patent/CN114742957A/en
Application granted granted Critical
Publication of CN114742957B publication Critical patent/CN114742957B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T15/00—3D [Three Dimensional] image rendering
    • G06T15/005—General purpose rendering architectures
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A30/00—Adapting or protecting infrastructure or their operation
    • Y02A30/60—Planning or developing urban green infrastructure

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Computer Graphics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Geometry (AREA)
  • Software Systems (AREA)
  • Processing Or Creating Images (AREA)

Abstract

The invention discloses a building facade extraction method based on point cloud data, which comprises the steps of preprocessing three-dimensional point cloud data of a facade to be extracted; creating a counter of 3D Hough transform based on the three-dimensional point cloud data, carrying out 3D high-pass filtering on the counter, and obtaining all potential planes based on the filtered counter; creating a constraint condition to carry out facade constraint on the potential plane, and obtaining a plane meeting the constraint condition as a potential facade; refining the potential vertical face to remove non-vertical face point cloud; carrying out elevation boundary calibration on the refined elevation; and carrying out the elevation constraint again on the calibration result to obtain the building elevation point cloud. The invention aims to provide a building facade extraction method based on point cloud data, and solves the technical problems of peak blurring, facade boundary confusion and the like existing when 3D Hough transform is used for building facade extraction in the prior art, so that the aims of weakening the influence of peak blurring, effectively distinguishing facade boundaries and the like are fulfilled.

Description

Building facade extraction method based on point cloud data
Technical Field
The invention relates to the field of image algorithms of building facades, in particular to a building facade extraction method based on point cloud data.
Background
Buildings are a major type of man-made object in urban market scenes. With the continuous development of city planning, smart cities and Building Information Models (BIMs), the demand of these fields for building structure information and characteristics thereof is increasing, and the demand is also increasing, and how to efficiently and accurately acquire these data and information required by 3D modeling is one of the main problems facing at present. The building elevation map reflects the geometric structure and the characteristics of the surface of a building, can directly serve old city reconstruction, city planning, smart city construction and the like, and provides a simple and flexible way for reconstructing a large-scale 3D building model. The cost is high and the efficiency is low when the traditional measuring method is used for collecting the building elevation, and the three-dimensional laser point cloud elevation extraction technology becomes a hot spot method for extracting the building elevation at present due to the unique advantages of the three-dimensional laser point cloud elevation extraction technology.
In the prior art, the inventor provides a 'constrained building facade orthophoto map extraction method' (with an authorization notice number CN 113256813B), the method overcomes the defect that the traditional technology depends on manual segmentation and input, and can realize full-automatic point cloud segmentation and full-automatic plane acquisition from the point cloud. However, the inventor finds that the method has the following disadvantages in the practical application process in a large amount of intensive research processes: (1) when the three-dimensional point cloud data is extracted through 3D Hough transformation, the balance between precision and efficiency cannot be achieved, the arithmetic operation amount and the memory overhead are changed into 4 times when the angular resolution is improved by 1 time, and the processing of a large amount of point cloud data is not friendly; (2) due to the reasons of step length setting, data noise and the like, the characteristic curves are usually not strictly intersected at one point, so that the problem of peak value blurring occurs in 3D Hough transformation, and great difficulty exists in facade extraction; (3) for the building point cloud data, due to the close proximity characteristic between buildings, the problem that the boundaries of adjacent and similar facades are difficult to distinguish exists, namely the problem that similar facades of different buildings are mistaken for the same plane easily exists. No relevant research in the prior art has appreciated the existence of the above drawbacks, let alone to address them.
Disclosure of Invention
The invention aims to provide a building facade extraction method based on point cloud data, and solves the technical problems of peak blurring, facade boundary confusion and the like existing when 3D Hough transform is used for building facade extraction in the prior art, so that the aims of weakening the influence of peak blurring, effectively distinguishing facade boundaries and the like are fulfilled.
The invention is realized by the following technical scheme:
a building facade extraction method based on point cloud data comprises the following steps:
preprocessing three-dimensional point cloud data of a facade to be extracted;
creating a counter of 3D Hough transform based on the three-dimensional point cloud data, carrying out 3D high-pass filtering on the counter, and obtaining all potential planes based on the filtered counter;
creating a constraint condition to carry out facade constraint on the potential plane, and obtaining a plane meeting the constraint condition as a potential facade;
refining the potential facade to remove non-facade point cloud;
carrying out elevation boundary calibration on the refined elevation;
and carrying out the elevation constraint again on the calibration result to obtain the building elevation point cloud.
Aiming at the problems of fuzzy peak value, confusion of facade boundary and the like in the process of extracting the facade of the building in the prior art, the invention provides a method for extracting the facade of the building based on point cloud data.
The inventor finds that, in the 3D hough transform, because a point in a cartesian coordinate system corresponds to a curve in a feature space, the corresponding feature curve theoretically representing the same straight line should intersect at a point, but due to step length setting, data noise and other reasons, the lines usually do not intersect at a point strictly, which causes a fuzzy phenomenon of a counter peak, and thus, the subsequent building facade extraction is difficult. To overcome this problem, the inventors performed 3D high-pass filtering on the created counters, which aims to remove low frequency parts in the counters to significantly attenuate the effect of peak blurring on building facade extraction.
The method adopts a counter after 3D high-pass filtering to complete a 3D Hough transformation process to obtain potential planes, then performs facade constraint on each potential plane, and defines the plane meeting the facade constraint condition as a potential facade; and then refining each potential vertical face, wherein the purpose of refining is to remove the same vertical face and a pseudo plane, and then performing vertical face boundary calibration on the refined vertical face.
For the building point cloud data, due to the characteristic of the close proximity between each building and adjacent buildings in modern city construction, the problem that the adjacent and similar vertical face boundaries are difficult to distinguish exists, in order to overcome the problem, the method introduces a vertical face boundary calibration process, and performs vertical face point cloud division through the vertical face boundary calibration, so that the building vertical face with a clear boundary can be obtained, and the effective distinction between the adjacent buildings is ensured. According to the method, through elevation boundary calibration, a plane (without a definite boundary range) in the mathematical sense is converted into an actual building elevation with a definite boundary, the identification and distinguishing capabilities of an opposite face and other ground object point clouds and point cloud sets close to different elevations are improved, the building elevation extraction precision is obviously improved, the influence of poor quality of point cloud data can be overcome, and different elevations are prevented from being combined into one elevation by mistake.
In addition, the inventor also finds that in the process of facade boundary calibration, cluster point clouds are separated easily due to clustering operation, namely one cluster is divided into a plurality of clusters, so that the facade constraint result is invalid, so that the cluster meeting the condition needs to be subjected to facade constraint again, namely the calibration result is subjected to facade constraint again, and a facade meeting the facade constraint condition is taken as the finally obtained building facade point cloud.
Therefore, the method and the device can not only fully automatically perform point cloud segmentation and fully automatically acquire the facade from the point cloud, but also overcome the problems of fuzzy characteristic curve peak value, confusion of building facade boundary and the like in the prior art, and remarkably improve the stability and accuracy of building facade extraction.
Further, the method for preprocessing the three-dimensional point cloud data of the vertical face to be extracted comprises the following steps:
removing ground points in the three-dimensional point cloud data;
translating the point cloud to a coordinate origin and performing voxel downsampling;
and removing statistical outliers of the data subjected to voxel down-sampling.
Because the three-dimensional point cloud data of the building comprises a large number of high-density ground points, the preprocessing of the scheme firstly eliminates the ground points; then, in order to reduce the subsequent calculation amount, the point cloud is translated to the coordinate origin and voxel down sampling (voxel down sampling) is carried out; and finally, performing statistical outlier removal (statistical outlier removal) on the data after the down sampling to remove noise point cloud.
Further, the method for creating the counter of the 3D Hough transform based on the three-dimensional point cloud data comprises the following steps:
dispersing the three-dimensional point cloud data through 3D Hough transformation to obtain a setM、N、Q(ii) a WhereinMAs a parameterθA discretized set of (a);Nas a parameterφA discretized set of (a);Qas a parameterρA discretized set of (a);θthe included angle between the normal vector of the plane of the point cloud and the z axis is set;φthe included angle between the plane normal vector of the point cloud and the x axis is set;ρthe distance from the origin to the point cloud plane;
pair setM、NPerforming offset counting to obtain offset copyM’、N’；
Separately creating countersA、A’：A=M×N×Q；A’ = M’×N’×Q。
The inventor finds that the largest challenge in the step length selection when the 3D Hough transform is applied to point cloud plane detection in a large amount of intensive research processes, and mainly shows that the discretization step length has great influence on the extraction process and the result: the smaller step length can generally obtain higher plane extraction precision, but the algorithm computation amount and the memory overhead are changed into 4 times of the original 4 times for each increase of the angular resolution (discretization step length), which is very unfriendly to the processing of the three-dimensional point cloud data of a large number of buildings, namely, the balance between precision and efficiency cannot be achieved. Therefore, the scheme optimizes the counter of the traditional 3D Hough transform, and the core of the scheme is that an offset copy is obtained by means of offset discretization step length and offset countingM’、N’Thus in a conventional counterABesides, another counter is obtainedA’All potential planes finally obtained are based on countersA、A’Respectively carrying out 3D high-pass filtering.
The counter establishing method improves the angle resolution, reduces the operation amount and the memory overhead, realizes the balance of precision and efficiency when the 3D Hough transform is applied to the building facade extraction, relieves the dependence of the traditional 3D Hough transform algorithm on step length selection, ensures that the precision is not limited by the step length selection any more, and obviously improves the precision and the robustness of the building facade extraction.
Further, the offset copyM’、N’Respectively are offset bys θ /2、s φ 2; whereins θ Is a setMThe discretization step length of (a);s φ is a setNIs used to determine the discretized step size.
That is to say, the number of the first and second,M’is a setMOffset of each element ins θ A copy of the/2-bit sequence,N’is a setNOffset of each element ins φ A copy of/2. The optimal offset mode can improve the angular resolution by 1 time through quantitative phase change, the calculation amount is only 2 times of the original calculation amount, and the memory overhead is the same as the original memory overhead, so that the balance between the precision and the efficiency when the 3D Hough transform is applied to the building facade extraction can be realized to the maximum extent.
Further, the method for obtaining all potential planes based on the filtered counters includes:
voting mechanism based on 3D Hough transform for filtered countersA、A’Voting to obtain a plane setSAndS’；
obtainingSAndS’as the potential plane.
The voting mechanism is the prior art in 3D hough transform, and is not described herein. The scheme obtains the counter through the stepsA、A’Then, firstly, make a pairA、A’Respectively carrying out 3D high-pass filtering and then carrying out 3D Hough transform pairA、A’Voting respectively, and obtaining results which are respectively plane setsS、S’(ii) a ObtainingSAndS’union ofS∪S’The final required potential plane can be obtained.
Further, in the 3D high-pass filtering: the value of the central cell of the convolution kernel is 1/2, and the values of the other cells are determined according to the distance from the central cell based on an inverse distance weighting method; the sum of the values of all cells is 1. The convolution kernel of the high-pass filtering is set in the optimal mode, so that the low-frequency part in the accumulator can be fully removed, and the obvious weakening effect on the characteristic curve peak value blurring phenomenon in the 3D Hough transform is further ensured.
Further, the facade constraint method comprises the following steps: and firstly carrying out coplanar constraint and then carrying out vertical plane constraint on the plane constrained by the coplanar constraint.
The constraint conditions of the coplanar constraint include: for any two planesp 1 、p 2 If the following formula is satisfied, it willp 1 Andp 2 merging as the same plane:
Figure 373571DEST_PATH_IMAGE002
in the formula (I), the compound is shown in the specification,r 12 respectively at the origin in the planep 1 、p 2 Distance vector between the upper drop legs;n 1 、n 2 are respectively a planep 1 、p 2 The plane normal vector of (a);ComPropan operator for solving the common point proportion of the two planes;α th is a plane dihedral angle between two planes; deltad th Is the distance between two planes;cp th a threshold corresponding to the common point proportion between the two planes;maxto solve the maximum operator; Λ is a logical or operator; the V.V.is a logical and operator;
the coplanar constraint aims at further removing the problems of the same plane and a pseudo plane caused by overlarge point cloud density, improper algorithm threshold setting and peak blurring, the coplanar constraint condition in the scheme is determined by three characteristics of a plane dihedral angle, a plane distance and a common point proportion, and compared with a mode of directly calculating the included angle of two planes in the prior art, the scheme reduces the possibility that adjacent real planes are combined by mistake by combining the two constraint conditions of the plane dihedral angle and the plane distance on one hand; on the other hand, by increasing the proportion constraint of the common points, the identification capability of the adjacent pseudo-planes is improved. Compared with the prior art, the scheme makes great progress in the accuracy of coplanar constraint.
The constraint conditions of the vertical plane constraint include: if the current plane meets the following formula, the current plane is regarded as a facade to be reserved, otherwise, the current plane is discarded:
Figure 831097DEST_PATH_IMAGE003
；
in the formula (I), the compound is shown in the specification,ma plane normal vector of a current plane;na plane normal vector of a vertical plane;α v,th the threshold is constrained for the vertical plane.
The plane extracted by the application not only comprises a building facade, but also comprises a plurality of other planes. Typically, the building facade should be a vertical plane, so constraining the vertical angle of each plane after being constrained to the same plane may exclude non-building facades. The scheme has the constraint condition of the vertical surface.
Further, the method for refining the potential facade comprises the following steps:
iterating each potential vertical face, and acquiring corresponding point clouds of the potential vertical faces by using a random sampling consistency algorithm;
clustering the corresponding point clouds of all the latent vertical surfaces by using an HDBSCAN algorithm to obtain a plurality of first clustered point clouds;
acquiring a plane equation of potential vertical faces corresponding to the first cluster point clouds by using a random sampling consistency algorithm, and taking the result as a new potential vertical face;
and (5) constraining the new potential vertical face, removing the same vertical face and the pseudo-plane, and finishing refining.
In addition, the inventor of the present application also finds that, in the research process, because the precision of facade extraction based on 3D hough transform is proportional to the step length, and the step length cannot be set to be very small under the normal circumstances in consideration of the calculation amount and the memory overhead, the absolute precision of facade extraction is not high, and non-facade point clouds (such as surface object point clouds of trees, street lamps, vehicles, and the like) near the facade are easily mistaken for the facade point clouds, so that the effect of facade refining in the present application is to remove the non-facade point clouds as much as possible.
The facade refining of the scheme integrates a RANSAC algorithm (random sampling consensus algorithm) and an HDBSCAN algorithm (Hierarchical Density-based Spatial Clustering of Applications with Noise, general terms, no standard Chinese translation, and direct translation into a 'Clustering algorithm based on Hierarchical Density with Noise'), has the characteristics of small influence by Noise, no step limit and the like, can extract a high-precision facade from a large number of potential facades, and overcomes the interference of non-facade point clouds near the facade. After a new potential vertical face is obtained, constraint is carried out again, and the same vertical face and the pseudo-plane are removed by setting corresponding constraint conditions.
Preferably, when the new potential facade is constrained in the facade refining process, the constraint conditions of the new potential facade can also adopt the constraint conditions of the 'facade constraint' mentioned above.
Further, the method for performing elevation boundary calibration on the refined elevation comprises the following steps:
clustering the refined facade point clouds by using an HDBSCAN algorithm to obtain a plurality of second cluster point clouds;
and respectively using a random sampling consistency algorithm for each second cluster point cloud, extracting an elevation equation and a corresponding elevation point cloud, and taking the minimum bounding box of the extracted elevation point cloud as an elevation boundary.
The purpose of the elevation boundary calibration is to overcome the problem that two elevations which are not coplanar but are adjacent in spatial position are regarded as one elevation. The inventor finds that, in the research process, for a vertical face with adjacent spatial positions, the three-dimensional point cloud data has the following characteristics: the density of point clouds in the vertical face of the building is usually higher, and the density of point clouds in the gaps between the vertical faces is lower, so that the density clustering idea of the HDBSCAN algorithm is very fit, and the HDBSCAN algorithm is introduced into the scheme and matched with the RANSAC algorithm (random sampling consensus algorithm) to overcome the problems.
The HDBSCAN algorithm is used as an advanced clustering algorithm, the application of the algorithm in the field of facade extraction is not available, and the algorithm can stably and efficiently perform clustering on large-scale data based on hierarchical density.
According to the scheme, after the refined facade point clouds are clustered by using the HDBSCAN algorithm, the RANSAC algorithm is adopted to extract a facade equation and the corresponding facade point clouds from the clustering result, and then the minimum bounding box of the extracted facade point clouds is found out and is used as a facade boundary, so that the recognition and distinguishing capabilities of facades and other ground object point clouds and near different facade point cloud sets are improved, the building facade extraction precision is further improved, the influence of poor data quality is overcome, and the error that different facades are mistakenly combined into a facade is avoided.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the building facade extraction method based on the point cloud data can not only carry out point cloud segmentation in a full-automatic manner and obtain facades from the point cloud in a full-automatic manner, but also overcome the problems that the counter peak value is fuzzy, the precision and the efficiency cannot be balanced, the building facade boundary is confused and the like in the prior art, and the robustness and the accuracy of building facade extraction are obviously improved.
2. According to the method for extracting the building facade based on the point cloud data, the counter creation of the 3D Hough transform is optimized by means of offset counting, the angle resolution is improved by phase change, the calculation amount and the memory overhead are reduced, the balance of precision and efficiency when the 3D Hough transform is applied to the building facade extraction is realized, the dependence of the traditional 3D Hough transform algorithm on step selection is relieved, the precision is not limited by the step selection any more, and the precision and the robustness of the building facade extraction are obviously improved.
3. The invention relates to a building facade extraction method based on point cloud data, which is characterized in that 3D high-pass filtering is carried out on a created counter in 3D Hough transform, and a low-frequency part in the counter is removed to obviously weaken the influence of peak value blurring on building facade extraction.
4. The invention discloses a building facade extraction method based on point cloud data, provides a brand-new facade refining process, and overcomes the interference of non-facade point clouds near a facade.
5. According to the building facade extraction method based on the point cloud data, a plane (without a definite boundary range) in a mathematical sense is converted into an actual building facade with a definite boundary through facade boundary calibration, the recognition and distinguishing capabilities of a facade and other ground object point clouds and point cloud sets close to different facades are improved, the building facade extraction precision is obviously improved, the influence of poor quality of the point cloud data can be overcome, and different facades are prevented from being combined into one facade by mistake.
Drawings
The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention and are incorporated in and constitute a part of this application, illustrate embodiment(s) of the invention and together with the description serve to explain the principles of the invention. In the drawings:
FIG. 1 is a schematic flow chart of an embodiment of the present invention;
FIG. 2 is a schematic diagram of convolution kernels for 3D high-pass filtering in an embodiment of the present invention;
FIG. 3 is a data set representation of point cloud data selected in an embodiment of the present invention;
FIG. 4 is a comparison graph of the effect of extracting a vertical plane by three different methods according to an embodiment of the present invention;
FIG. 5 is a diagram of an error comparison violin in an embodiment of the present disclosure;
FIG. 6 is a histogram of error contrast in an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail below with reference to examples and accompanying drawings, and the exemplary embodiments and descriptions thereof are only used for explaining the present invention and are not meant to limit the present invention.
Example 1:
fig. 1 shows a building facade extraction method based on point cloud data, which includes the following steps:
preprocessing three-dimensional point cloud data of a facade to be extracted;
establishing a counter of 3D Hough transform based on the three-dimensional point cloud data, carrying out 3D high-pass filtering on the counter, and obtaining all potential planes based on the filtered counter;
creating a constraint condition to carry out facade constraint on the potential plane, and obtaining a plane meeting the constraint condition as a potential facade;
refining the potential facade to remove non-facade point cloud;
carrying out elevation boundary calibration on the refined elevation;
and carrying out the elevation constraint again on the calibration result to obtain the point cloud of the elevation of the building.
The counter creating method comprises the following steps:
dispersing the three-dimensional point cloud data through 3D Hough transformation to obtain a setM、N、Q(ii) a WhereinMAs a parameterθA discretized set of (a);Nas a parameterφThe discretized set of (a);Qas a parameterρA discretized set of (a);θthe included angle between the normal vector of the plane of the point cloud and the z axis is set;φthe included angle between the plane normal vector of the point cloud and the x axis is set;ρthe distance from the origin to the point cloud plane;
pair setM、NPerforming offset counting to obtain offset copyM’、N’；
Separately creating countersA、A’：A=M×N×Q；A’ = M’×N’×Q。
Wherein the copy is offsetM’、N’Are respectively offset bys θ /2、s φ 2; whereins θ Is a setMThe discretization step length of (a);s φ is a setNThe discretization step size of (a).
The method for obtaining all potential planes based on the filtered counters comprises the following steps:
voting mechanism based on 3D Hough transform for filtered countersA、A’Voting to obtain a plane setSAndS’；
obtainingSAndS’as the potential plane.
Example 2:
a building facade extraction method based on point cloud data mainly comprises the following steps:
1. and (4) point cloud preprocessing.
Because the point cloud data contains a large number of ground points and has high density, the ground points are removed firstly. In addition, in order to reduce the amount of computation, the point cloud is translated to the origin of coordinates and voxel down sampling (voxel down sampling) is performed. Finally, statistical outlier removal (statistical outlier removal) is performed on the down-sampled data to remove noise point clouds.
2. Extraction of potential planes by improved 3D Hough transform
(1) Creating a counter and an offset count
When 3D Hough transform plane parameter discretization is carried out,ρdiscretized into collections according to the prior artQ，θAndφin the prior art discretization into setsM{0, s θ ,2s θ ,…,2πAndN{0, s φ ,2 s φ ,…,2πafter is againM、NRespectively creating an offset for each elements θ /2、s φ A replica of/2, defined as an offset replicaM’、N’：
Figure 645469DEST_PATH_IMAGE004
Wherein the content of the first and second substances,θthe included angle between the plane normal vector of the point cloud and the z axis is set;φthe included angle between the plane normal vector of the point cloud and the x axis is set;ρis the distance from the origin to the point cloud plane.
On the basis of the above-mentioned data, the following counters are respectively createdA、A’：
Figure 467932DEST_PATH_IMAGE005
Wherein the content of the first and second substances,θ’is a pair ofOffset of theta s θ A replica element of/2;φ’is a pair ofφOffset ofs φ A replica element of/2; subscriptj、ijRespectively, the position of the elements.
(2) 3D high pass filtering
Respectively to counterA、A’3D high-pass filtering is carried out, and the low-frequency part in the counter is removed to weaken the influence of peak value blurring. The convolution kernel of the 3D high-pass filtering in this embodiment is shown in fig. 2, and each number in fig. 2 represents a value of each cell in the convolution kernel, it can be seen that a central cell value of the convolution kernel is 1/2, values of the remaining cells are determined according to distances from the central cell based on an inverse distance weighting principle, and a sum of values of the entire convolution kernels is 1.
(3) Potential plane acquisition
For the filtered counterA、A’Voting to obtain a candidate plane set satisfying the conditionS(θ, φ, ρ) AndS’(θ’, φ’, ρ) (ii) a Using a union of the twoS∪S’As the final set of candidate planes.
3. Facade restraint
The method is mainly realized through coplanar constraint and vertical plane constraint.
(1) Coplanar constraint
The same plane constraint aims to further remove the problems of the same plane and a pseudo plane caused by overlarge point cloud density, improper algorithm threshold setting and peak blurring, and is mainly determined by three characteristics of a plane dihedral angle, a plane distance and a common point proportion. If two planes arep 1 、p 2 Satisfying the following formula, willp 1 、p 2 Treated as the same plane and merged:
Figure 734965DEST_PATH_IMAGE007
in the formula (I), the compound is shown in the specification,r 12 respectively at the origin in the planep 1 、p 2 Distance vector between the upper drop feet;n 1 、n 2 are respectively a planep 1 、p 2 A plane normal vector of (a);ComPropan operator for solving the proportion of the common points of the two planes;α th is a plane dihedral angle between two planes; deltad th Is the distance between the two planes;cp th a threshold value corresponding to the common point proportion between the two planes;maxto solve for the maximum operator; Λ is a logical or operator; v is a logic and operator;
(2) vertical plane constraint
The potential plane includes not only the facade of the building but also many other planes. In general, a building facade should be a vertical plane, so that the constraint on the vertical angles of planes after constraint on the same plane can exclude non-building facades, and the constraint conditions of the vertical plane constraint are as follows:
Figure 996182DEST_PATH_IMAGE003
；
in the formula (I), the compound is shown in the specification,ma plane normal vector of the current plane;na plane normal vector of a vertical plane;α v,th the threshold is constrained for the vertical plane.
If the potential plane does not meet the vertical plane constraint, the potential plane is directly discarded.
And obtaining the potential vertical face after coplanar constraint and vertical plane constraint.
4. Elevation refining
(1) Iterating each potential facade, and acquiring point clouds corresponding to the potential facades by using a RANSAC algorithm;
(2) clustering the point clouds of all the potential vertical surfaces by using an HDBSCAN algorithm to obtain point cloud clusters of all the points;
(3) acquiring a plane equation of a potential vertical face corresponding to each point cloud cluster by using a RANSAC algorithm, and taking a result as a new potential vertical face;
(4) and (5) potential facade constraint, removing the same facade and the pseudo plane to obtain the refined building facade.
It should be noted that RAN is used because the present embodiment is directed to facade extraction, and the building facade is generally a vertical surfaceWhen the SAC algorithm carries out plane extraction, the plane equation of the algorithm is orderedAx+By+Cz+DIn =0C=0, the plane is constrained to be a vertical plane to improve the plane extraction accuracy.
5. Elevation boundary calibration
Clustering the point clouds after the facade refining by using an advanced HDBSCAN algorithm, then extracting an facade equation and corresponding facade point clouds from the generated clustered point clouds by using a RANSAC algorithm respectively, and taking a minimum bounding box of the facade point clouds as a facade boundary.
Because clustering of the HDBSCAN algorithm causes that each clustered point cloud can be separated, and a cluster can be divided into a plurality of clusters, the elevation constraint recorded in the step 3 needs to be executed again on the result of elevation boundary calibration, so that the building elevation with the boundary and the corresponding point cloud data thereof are obtained.
Example 3:
in this embodiment, the building facade extraction method described in embodiment 2 is used, and the validity of the method is verified by substituting the method into an example.
This example was performed on the IQmulus & terrnamobilia content dataset. The IQmulus & terra Mobile content dataset is public 3D MLS data of a paris dense urban environment in france collected by Mobile Laser Scanning (MLS) consisting of 3 hundred million points. Due to the limitations of current laser scanning techniques, the data set has problems such as density non-uniformity, noise, occlusion, etc.; in addition, the data set has the problem of splicing dislocation in some areas, so that the data set is suitable for the verification of the method.
This embodiment selects a subset of the data set as experimental data, and the range of the subset data is shown by the dashed box in fig. 3.
The IQmulus & Terra Mobile content data set is divided into 9 data files, and the files are firstly combined into one file and cut out to obtain the experimental data. Meanwhile, the extraction algorithm only needs the space coordinate information of the point cloud, and other attributes (such as reflectivity, echo times and the like) except the coordinate information are removed, so that the data volume is reduced. Thereafter, the data was preprocessed by the point cloud preprocessing method described in example 2, and the processed data consisted of 2300 ten thousand points. And then extracting the building facade from the processed point cloud data by using the building facade extraction method in the embodiment 2.
As a comparative example, the conventional 3D RANSAC algorithm (the General iterative RANSAC method, abbreviated as the GIR method) and the 3D RANSAC algorithm with the elevation constraint (the Vertical constrained RANSAC method, abbreviated as the VCIR method) were used together to perform the elevation extraction, and the overall extraction effect of the 3 methods is shown in fig. 4, and the upper and lower rows in fig. 4 respectively show the effect of the three extracted elevations at the 3D viewing angle and the 2D viewing angle.
As can be seen from fig. 4, the GIR method is completely inapplicable to this scenario, and the mentioned "facades" are all horizontal planes, and the correct building facades cannot be extracted; both the VCIR method and the method proposed by the present invention are able to extract most facades from experimental data. This example therefore continues to compare the results of the inventive method and the VCIR method.
As can be seen from FIG. 4, the adjacent or similar vertical faces are regarded as the same vertical face by the VCIR method, but the adjacent or similar vertical faces are divided by the method of vertical face boundary calibration, so that the vertical faces are correctly distinguished from each other.
In addition, in order to better compare the advantages of the method of the present invention compared to the VCIR method, the extracted vertical surfaces of the two are numbered by using roman numerals and capital english letters respectively in the 2D view of fig. 4, and the specific numbering is shown in fig. 4, and then table 1 is counted based on the numbering.
TABLE 1 comparison of extraction results
Figure 665061DEST_PATH_IMAGE008
As can be taken from table 1 in conjunction with fig. 4, one facade extracted by the VCIR process is generally treated by the present process as a plurality of mutually independent facades. Compared with the traditional method, the extraction result of the method has higher accuracy and reliability.
Further, in order to quantitatively evaluate the effects of the method and the VCIR method, the distance from the point cloud corresponding to the facade is taken as an error, and each facade error and the total error obtained by the two methods are respectively calculated, and the results are shown in Table 2.
TABLE 2 vertical extraction error Table
Figure 658424DEST_PATH_IMAGE009
As can be seen from Table 2, in the total errors extracted from the facade by the method, the average Absolute error MAE (mean Absolute error) and the mean Square error MSE (mean Square error) are respectively 0.314m and 0.194m, and are respectively 62.8 percent and 48.1 percent of the corresponding total errors of 0.500m and 0.403m of the VCIR method. In addition, the minimum mean square error MSE corresponding to the VCIR method is 0.208m, and the average value is as high as 0.439m, while the minimum mean square error MSE in each facade extracted by the method is only 0.085m, and the average value is only 0.271m, which are both about half of the VCIR method. Therefore, compared with the VCIR method, the facade extraction method can obviously reduce errors caused by density unevenness, noise, shading, dislocation and the like.
Furthermore, in order to further evaluate the effect of the method of the present invention and the VCIR method, a violin chart and a bar chart are respectively used to show the distributions of the error values MAE, MSE and the root Mean Square error rmse (root Mean Square error) of the planes obtained by the two methods, as shown in fig. 5 and fig. 6.
As can be seen from FIG. 5, the median and quartile of the three errors of the extracted result of the method of the present invention are both smaller than those of the VCIR method, and meanwhile, the probability density distribution of the method is spindle-shaped, narrow at the top and wide at the bottom, and the overall level of the errors is low. The VCIR method has the advantages that the probability density distribution is in a gourd shape integrally, and the error distribution is uniform. As can be seen from FIG. 6, the average of the three errors extracted by the method of the present invention is smaller than that of the VCIR method, and the error distribution is more concentrated and stable.
In conclusion, the building facade extraction method based on the point cloud data has obvious advantages compared with the prior art in all aspects of robustness, extraction precision, extraction details and the like.
It should be noted that, in this document, relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
The above-mentioned embodiments are intended to illustrate the objects, technical solutions and advantages of the present invention in further detail, and it should be understood that the above-mentioned embodiments are merely exemplary embodiments of the present invention, and are not intended to limit the scope of the present invention, and any modifications, equivalent substitutions, improvements and the like made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims (7)

1. A building facade extraction method based on point cloud data is characterized by comprising the following steps:
preprocessing three-dimensional point cloud data of a facade to be extracted;
creating a counter of 3D Hough transform based on the three-dimensional point cloud data, carrying out 3D high-pass filtering on the counter, and obtaining all potential planes based on the filtered counter;
creating a constraint condition to carry out facade constraint on the potential plane, and obtaining a plane meeting the constraint condition as a potential facade;
refining the potential facade to remove non-facade point cloud;
carrying out elevation boundary calibration on the refined elevation;
performing the elevation constraint on the calibration result again to obtain a building elevation point cloud;
wherein the content of the first and second substances,
the method for creating the counter of the 3D Hough transform based on the three-dimensional point cloud data comprises the following steps:
dispersing the three-dimensional point cloud data through 3D Hough transformation to obtain a setM、N、Q(ii) a WhereinMAs a parameterθA discretized set of (a);Nas a parameterφA discretized set of (a);Qas a parameterρThe discretized set of (a);θthe included angle between the normal vector of the plane of the point cloud and the z axis is set;φthe included angle between the plane normal vector of the point cloud and the x axis is set;ρthe distance from the origin to the point cloud plane;
to the collectionM、NPerforming offset counting to obtain offset copyM’、N’；
Separately creating countersA、A’：A=M×N×Q；A’ = M’×N’×Q；
In the 3D high-pass filtering: the value of the central cell of the convolution kernel is 1/2, and the values of the other cells are determined according to the distance from the central cell based on an inverse distance weighting method; the sum of the values of all cells is 1;
the method for refining the potential facade comprises the following steps:
iterating each potential vertical face, and acquiring corresponding point clouds of the potential vertical faces by using a random sampling consistency algorithm;
clustering the corresponding point clouds of all the latent vertical surfaces by using an HDBSCAN algorithm to obtain a plurality of first clustered point clouds;
acquiring a plane equation of potential vertical faces corresponding to the first cluster point clouds by using a random sampling consistency algorithm, and taking the result as a new potential vertical face;
and (5) constraining the new potential vertical face, removing the same vertical face and the pseudo-plane, and finishing refining.
2. The method for extracting the facade of the building based on the point cloud data as claimed in claim 1, wherein the method for preprocessing the three-dimensional point cloud data of the facade to be extracted comprises the following steps:
removing ground points in the three-dimensional point cloud data;
translating the point cloud to a coordinate origin and performing voxel downsampling;
and removing statistical outliers of the data subjected to voxel downsampling.
3. The method of claim 1, wherein the offset copy is used for extracting the facade of the building based on the point cloud dataM’、N’Respectively are offset bys θ /2、s φ 2; whereins θ Is a setMThe discretization step length of (a);s φ is a setNIs used to determine the discretized step size.
4. The method for extracting the facade of the building based on the point cloud data as claimed in claim 1, wherein the method for obtaining all the potential planes based on the filtered counter comprises:
voting mechanism based on 3D Hough transform for filtered countersA、A’Voting to obtain a plane setSAndS’；
obtainingSAndS’as the union of potential planes.
5. The method for extracting the facade of the building based on the point cloud data as claimed in claim 1, wherein the facade constraint method comprises the following steps: and firstly carrying out coplanar constraint and then carrying out vertical plane constraint on the plane constrained by the coplanar constraint.
6. The method for extracting the facade of the building based on the point cloud data as claimed in claim 5,
the constraint conditions of the coplanar constraint include: for any two planesp 1 、p 2 If the following formula is satisfied, the following formula will be satisfiedp 1 Andp 2 merging as the same plane:
Figure DEST_PATH_IMAGE002
；
in the formula (I), the compound is shown in the specification,r 12 respectively at the plane of originp 1 、p 2 Distance vector between the upper drop legs;n 1 、n 2 are respectively a planep 1 、p 2 A plane normal vector of (a);ComPropan operator for solving the common point proportion of the two planes;α th is a plane dihedral angle between two planes; deltad th Is the distance between two planes;cp th a threshold corresponding to the common point proportion between the two planes;maxto solve the maximum operator; Λ is a logical or operator; v is a logic and operator;
the constraint conditions of the vertical plane constraint include: if the current plane meets the following formula, the current plane is regarded as a facade to be reserved, otherwise, the current plane is discarded:
Figure DEST_PATH_IMAGE004
；
in the formula (I), the compound is shown in the specification,ma plane normal vector of the current plane;na plane normal vector of a vertical plane;α v,th the threshold is constrained for the vertical plane.
7. The method for extracting the facade of the building based on the point cloud data as claimed in claim 1, wherein the method for calibrating the facade boundary of the refined facade comprises the following steps:
clustering the refined facade point clouds by using an HDBSCAN algorithm to obtain a plurality of second cluster point clouds;
and respectively using a random sampling consistency algorithm for each second cluster point cloud, extracting an elevation equation and the corresponding elevation point cloud, and taking the minimum bounding box of the extracted elevation point cloud as an elevation boundary.
CN202210659698.2A 2022-06-13 2022-06-13 Building facade extraction method based on point cloud data Active CN114742957B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202210659698.2A CN114742957B (en) 2022-06-13 2022-06-13 Building facade extraction method based on point cloud data

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202210659698.2A CN114742957B (en) 2022-06-13 2022-06-13 Building facade extraction method based on point cloud data

Publications (2)

Publication Number Publication Date
CN114742957A CN114742957A (en) 2022-07-12
CN114742957B true CN114742957B (en) 2022-08-19

Family

ID=82286963

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202210659698.2A Active CN114742957B (en) 2022-06-13 2022-06-13 Building facade extraction method based on point cloud data

Country Status (1)

Country Link
CN (1) CN114742957B (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116310197B (en) * 2023-05-11 2023-08-25 四川省非物质文化遗产保护中心 Three-dimensional model construction method, device and storage medium for non-genetic building
CN117745728B (en) * 2024-02-21 2024-05-24 法奥意威(苏州)机器人系统有限公司 Point cloud plane detection method and device, electronic equipment and readable storage medium

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110717983A (en) * 2019-09-07 2020-01-21 苏州工业园区测绘地理信息有限公司 Building facade three-dimensional reconstruction method based on knapsack type three-dimensional laser point cloud data

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113155032A (en) * 2020-12-16 2021-07-23 重庆邮电大学 Building structure displacement measurement method based on dynamic vision sensor DVS
CN113256813B (en) * 2021-07-01 2021-09-17 西南石油大学 Constrained building facade orthophoto map extraction method
CN114494292A (en) * 2022-01-19 2022-05-13 武汉大势智慧科技有限公司 Method and system for extracting building facade glass area

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110717983A (en) * 2019-09-07 2020-01-21 苏州工业园区测绘地理信息有限公司 Building facade three-dimensional reconstruction method based on knapsack type three-dimensional laser point cloud data

Also Published As

Publication number Publication date
CN114742957A (en) 2022-07-12

Similar Documents

Publication Publication Date Title
CN114742957B (en) Building facade extraction method based on point cloud data
Dick et al. Modelling and interpretation of architecture from several images
CN102804231B (en) Piecewise planar reconstruction of three-dimensional scenes
Xu et al. Reconstruction of scaffolds from a photogrammetric point cloud of construction sites using a novel 3D local feature descriptor
Canaz Sevgen et al. An improved RANSAC algorithm for extracting roof planes from airborne lidar data
Leng et al. A multi‐scale plane‐detection method based on the Hough transform and region growing
WO2022141145A1 (en) Object-oriented high-resolution remote sensing image multi-scale segmentation method and system
CN110047036B (en) Polar grid-based ground laser scanning data building facade extraction method
CN114742968B (en) Elevation map generation method based on building elevation point cloud
CN113538501A (en) Low-altitude image DSM generation building edge refinement method
Satari et al. A multi‐resolution hybrid approach for building model reconstruction from lidar data
Wei et al. Automatic coarse registration of point clouds using plane contour shape descriptor and topological graph voting
Zhu et al. Multiple close‐range image matching based on a self‐adaptive triangle constraint
CN110580705B (en) Method for detecting building edge points based on double-domain image signal filtering
CN108388902A (en) Sub- construction method is described in conjunction with the compound 3D of global frame point and part SHOT features
Park et al. Segmentation of Lidar data using multilevel cube code
Peng et al. Incorporating generic and specific prior knowledge in a multiscale phase field model for road extraction from VHR images
WO2024125434A1 (en) Regional-consistency-based building principal angle correction method
Ebrahimikia et al. True orthophoto generation based on unmanned aerial vehicle images using reconstructed edge points
CN113836484A (en) Self-adaptive point cloud rarefying method based on path point adjacent domain and ground filtering
CN113435479A (en) Feature point matching method and system based on regional feature expression constraint
Yan et al. Quality analysis on ransac-based roof facets extraction from airborne lidar data
CN115131459B (en) Reconstruction method and device for floor plan
CN114463396B (en) Point cloud registration method utilizing plane shape and topological graph voting
Liu et al. The classification of airborne LiDAR building point clouds based on multi‐scale and multi‐level cloth simulation

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant