CN106780524B - Automatic extraction method for three-dimensional point cloud road boundary - Google Patents

Automatic extraction method for three-dimensional point cloud road boundary Download PDF

Info

Publication number
CN106780524B
CN106780524B CN201610996779.6A CN201610996779A CN106780524B CN 106780524 B CN106780524 B CN 106780524B CN 201610996779 A CN201610996779 A CN 201610996779A CN 106780524 B CN106780524 B CN 106780524B
Authority
CN
China
Prior art keywords
point
points
road boundary
hyper
boundary points
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
CN201610996779.6A
Other languages
Chinese (zh)
Other versions
CN106780524A (en
Inventor
李军
宰大卫
林阳斌
郭裕兰
王程
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.)
Xiamen University
National University of Defense Technology
Original Assignee
Xiamen 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 Xiamen University filed Critical Xiamen University
Priority to CN201610996779.6A priority Critical patent/CN106780524B/en
Publication of CN106780524A publication Critical patent/CN106780524A/en
Application granted granted Critical
Publication of CN106780524B publication Critical patent/CN106780524B/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
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30248—Vehicle exterior or interior
    • G06T2207/30252—Vehicle exterior; Vicinity of vehicle
    • G06T2207/30256—Lane; Road marking

Landscapes

  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention relates to the field of point cloud processing, and particularly discloses a three-dimensional point cloud road boundary automatic extraction method which comprises the following steps of S1, screening seed points for a whole three-dimensional point cloud data set P to perform voxel division, S2, extracting boundary points between adjacent non-coplanar voxels by using a α -shape algorithm, S3, extracting road boundary points by using an energy minimization algorithm based on graph cut, S4, removing outliers based on an Euclidean distance clustering algorithm, and S5, fitting the extracted road boundary points into a smooth curve.

Description

Automatic extraction method for three-dimensional point cloud road boundary
Technical Field
The invention relates to the field of point cloud processing, in particular to a method for automatically extracting a three-dimensional point cloud road boundary.
Background
The road is used as a traffic infrastructure, and the digital management and construction of the road have important significance for application such as city planning, traffic management and navigation. As a high and new surveying and mapping technology which develops rapidly, compared with the traditional surveying and mapping means, the vehicle-mounted laser scanning technology has the advantages of high data acquisition speed, high data precision, non-contact active measurement, strong real-time performance and the like, can quickly acquire detailed three-dimensional space information of roads and ground objects on two sides in the normal traveling process of a vehicle, and has obvious advantages for acquiring road information in strip distribution.
The traditional method for acquiring road information mainly comprises two modes of manual measurement and digital photogrammetry. Although the manual measurement can obtain more accurate information such as road coordinates and the like, the measurement speed is slow, and the updating period of the road information is long; digital photogrammetry is gradually developed and matured along with the development of scientific technology and the wide application of high and new technologies such as computers, but is limited by reasons such as image resolution and the like, and the requirements on road characteristic information and precision extracted from images still need to be further improved. The vehicle-mounted laser scanning system comprises a global positioning system, an inertial navigation system, a laser scanner, a CCD camera and the like, and becomes a new means for acquiring three-dimensional space data. The vehicle-mounted laser scanning technology can effectively save the measurement time, improve the measurement efficiency, shorten the road information updating period, avoid the exposure risk of measurement operators in the traffic environment and provide powerful technical support for the survey and planning of urban space resources.
However, the city is usually complex in environment, not only are the accessory components complex and numerous, but also the scanned targets are shielded from each other to cause data loss, and the automatic extraction of the road boundary is examined. In addition, the complexity caused by different road environments (such as vehicle parking, vegetation surrounding, fences, etc.) increases the difficulty of automatic extraction of road boundaries. Therefore, the difficulty and the requirement for rapidly and automatically extracting the road boundary from the massive point cloud are high, but the technology has important economic and application requirements and is a research hotspot at home and abroad.
At present, most of research on vehicle-mounted laser scanning data processing focuses on ground object point cloud classification, building facade information extraction and modeling, road accessory facility extraction and the like, while research on road boundary information extraction is relatively few, and main work can be divided into indirect extraction and direct extraction.
The indirect extraction method generally first uses the attributes (height, intensity, wavelength, etc.) of the point cloud to generate a depth image, and then uses image processing methods (cropping, fitting, filtering, etc.) to detect and extract the road boundary. For indirect extraction methods, namely converting point clouds into depth images and then extracting road boundaries by using image processing methods, errors are inevitably generated in the conversion process by the methods, and accurate road boundary results are difficult to obtain.
The direct extraction method generally uses road features (such as planes, road teeth, etc.) to detect and extract road boundaries. A common method is to extract the road surface by a random sample consensus (RANSAC) method and then obtain the road boundary by a linear fitting algorithm. The road tooth is detected by a gaussian filtering method or a sliding window method, thereby obtaining a road boundary. For the direct extraction method, the application range of the scene is greatly limited. The method of extracting the road surface based on random sample consensus (RANSAC) has difficulty in the case of road undulation, and some details of the extracted road surface are lost. The method of using gaussian filtering or sliding window to detect the road tooth is often challenging when facing irregular road boundaries (such as walls, fences) or vegetation surrounding.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides the automatic extraction method of the three-dimensional point cloud road boundary, which can be directly operated on large-scale three-dimensional point cloud, can be used for different scenes, has high calculation speed and good algorithm robustness, and can quickly extract the road boundary.
The specific technical scheme is as follows:
a three-dimensional point cloud road boundary automatic extraction method comprises the following steps:
s1, screening seed points of the obtained whole three-dimensional point cloud data set P for superpixel division;
s2, extracting boundary points between adjacent non-coplanar hyper-voxels by using a α -shape algorithm;
s3, extracting road boundary points by using an energy minimization algorithm based on graph cut;
s4, removing outliers based on an Euclidean distance clustering algorithm;
and S5, fitting the extracted road boundary points into a smooth curve.
Preferably, the partitioning procedure of the super-voxel in step S1 is as follows:
s11 solving fitting plane Tp(pi)：
For each of the entire three-dimensional point cloud data sets PInput point piIts tangent plane Tp (p)i) Can be represented as a point o from its centeriSum normal vector nlFormed doublets, i.e.
Figure BDA0001150616040000031
Any point p to Tp (p) in three-dimensional spacei) Can be expressed as
Figure BDA0001150616040000032
Note piK neighbors of (a) are set of Nbk(pi) The best fit plane in the least squares sense can be obtained by solving the following equation
Figure BDA0001150616040000033
The fitted plane is then optimized using an iterative reweighted least squares method
Figure BDA0001150616040000034
Solving the least square equation of the band weight to obtain the optimized fitting plane Tp (p)i)
Figure BDA0001150616040000035
To plane Tp (p)i) The above process is repeated until the algorithm converges.
S12, removing non-ground points:
note that the final component tangent plane Tp (p)i) Three eigenvalues of the covariance matrix of the point set of (a) are λ1,λ2And λ3And satisfy lambda1≥λ2≥λ3. Then point piSmoothness of(s) (p)i) Can be expressed as
Figure BDA0001150616040000041
The following two constraints are used to remove non-ground points:
A. removing points significantly above the road surface (z)i≥5m)(ziIs a point piHeight value of (d);
B. and removing the point of which the included angle between the normal vector and the Z axis is more than 22.5 degrees.
S13, calculating the hyper-voxel fi：
Removing the point set P after the non-ground pointgAccording to the smoothness ranking of each point, the point with high smoothness is selected as a seed point. The hyper-voxels are computed in such a way that region growing starts from the seed point. Let voxel fiFormalized definition as a set of points P to which there belongsiCenter point oiAnd the normal vector nlFormed triplet
Figure BDA0001150616040000042
Seed for each seed pointiLet its initial hyper-voxel fiIs { p }iCenter point and normal vector are Tp (p) respectivelyi).oiAnd Tp (p)i).nl. Then, the principle of width priority is adopted to fiA region growing is performed. For each candidate point pjIf (1) p is satisfiedjTo piIs less than a threshold value Rseed(ii) a (2) Vector Tp (p)j).nlAnd Tp (p)i).nlThe included angle of the angle is less than 22.5 degrees; (3) p is a radical ofjTo Tp (p)i) Is less than a threshold e; then p will bejIs added to fiIs concentrated. When f isiWhen no longer expandable, according to fi.piFitting the plane using least squares, and fitting fi.nlg is updated as the normal vector of the fitted plane. And (3) assigning points to the hyper-voxels by adopting local K-means clustering on the basis of the initial facets, and ensuring that the distance from each point to the hyper-voxel to which the point belongs is smaller than the distance from each point to other hyper-voxels. The distance function here is defined as:
Figure BDA0001150616040000043
wherein Ds,DnAnd DiRespectively euclidean distance, normal vector distance and intensity distance. Omegas,ωnAnd ωiRespectively, the corresponding weights.
Preferably, the specific steps of extracting boundary points between adjacent non-coplanar hyper-voxels using the α -shape algorithm in step S2 are as follows:
after the point cloud is segmented into the hyper-voxels, for each hyper-voxel, an α -shape algorithm can be used to extract boundary points, and meanwhile, the boundary points between two hyper-voxels coplanar with each other are removed, namely, if the included angle of normal vectors of the two hyper-voxels is less than 22.5 degrees, the boundary points between the two hyper-voxels are deleted, and the boundary point P is the boundary point at the momentbIncluding road boundary points and non-road boundary points.
Preferably, step S3 uses an energy minimization algorithm based on graph cut to extract road boundary points, as follows:
and providing vehicle running track line data by a vehicle-mounted laser scanning system, and taking the track line data as an initial observation model of a graph cutting algorithm. The energy formula is defined as:
E(f)＝Edata(f)+λ·Esmooth(f)
Figure BDA0001150616040000051
Figure BDA0001150616040000052
Figure BDA0001150616040000053
where P isbRefers to the set of boundary points extracted in step 2. n is piThe potential of the point set of the belonging superpixel. Δ djIs pointing at pjTo a straight line LpiThe distance of (c). Δ diIs pointing at piAll points in the neighborhood to the straight line LpiAverage redundancy of (2). Sigma1Mean redundancy of all points. Straight line LpiIs defined as passing through point piAnd the direction and the distance piThe direction of the nearest trajectory line.
Figure BDA0001150616040000054
Figure BDA0001150616040000055
Figure BDA0001150616040000056
Figure BDA0001150616040000057
(xi,yi,zi),(xj,yj,zj) Are respectively a point piAnd pjIs determined by the three-dimensional coordinates of (a),
Figure BDA0001150616040000058
is pointing at piAnd pjThe euclidean distance of (c). Here, the
Figure BDA0001150616040000059
Indicates if the point piAnd pjIf the allocated labels are consistent, the cost is zero, otherwise, the cost is
Figure BDA0001150616040000061
Where σ is2Refers to a set of points PbThe spatial resolution of (a). The result of using the graph cut algorithm to find the minimum value of the above energy formula is to divide the boundary points into two types, one is the road boundary points, and the other is the non-road boundary points.
Compared with the prior art, the scheme of the invention has the following advantages:
(1) the method can be directly operated on large-scale three-dimensional point cloud, and provides a set of rapid, effective and automatic solution for extracting and positioning the road boundary. The parameters which need to be set manually are very few, and the subjective intervention of human is reduced. Compared with the prior art, the method adopts the voxel segmentation and the minimum energy algorithm based on the graph segmentation, can still effectively extract the road boundary under the complex urban environment condition, overcomes the defects of point cloud data shielding, uneven density distribution and the like due to the combined use of vehicle-mounted system trajectory data for calculation, enables the result to be stable and robust, has universality for different scenes, and is easy to practically apply.
(2) The method fully excavates the basic attributes (spatial distance, geometric properties and intensity information) of the point cloud, carries out hyper-voxel segmentation on the point cloud, removes non-ground points and improves the subsequent calculation efficiency. Due to the method of sequencing and preferentially selecting the seed points, the boundary information is well preserved in the super-voxel segmentation result, and the robustness of the subsequent road boundary extraction algorithm is improved.
(3) The method is characterized in that an energy minimization algorithm based on graph cut is provided for the first time, a graph cut model is established by utilizing trajectory line data provided by a vehicle-mounted system and combining the internal characteristics of the road boundary, and the road boundary is effectively and quickly extracted.
Drawings
FIG. 1 is a schematic flow diagram of the present invention;
FIG. 2 illustrates raw point cloud data according to an embodiment of the present invention;
figure 3 shows the effect after treatment.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Examples
The specific implementation of the method for extracting the road boundary based on the vehicle-mounted laser scanning point cloud data provided by the invention is as follows (the flow of the general technical scheme can be shown in figure 1):
s1, screening seed points for the obtained whole three-dimensional point cloud data set P to perform voxel division (the original point cloud data in this embodiment can be shown in fig. 2);
the hyper-voxel division refers to gathering adjacent points with consistent properties into a hyper-point so as to reduce the complexity of data processing;
s11 solving fitting plane Tp(pi)：
For each input point P of the entire three-dimensional point cloud dataset PiIts tangent plane Tp (p)i) Can be represented as a point o from its centeriSum normal vector nlFormed doublets, i.e.
Figure BDA0001150616040000071
Any point p to Tp (p) in three-dimensional spacei) Can be expressed as
Figure BDA0001150616040000072
Note piK neighbors of (a) are set of Nbk(pi) The best fit plane in the least squares sense can be obtained by solving the following equation
Figure BDA0001150616040000073
The fitted plane is then optimized using an iterative reweighted least squares method
Figure BDA0001150616040000074
Solving the least square equation of the band weight to obtain the optimized fitting plane Tp (p)i)
Figure BDA0001150616040000075
To plane Tp (p)i) The above process is repeated until the algorithm converges.
S12, removing non-ground points:
note that the final component tangent plane Tp (p)i) Three eigenvalues of the covariance matrix of the point set of (a) are λ1,λ2And λ3And satisfy lambda1≥λ2≥λ3. Then point piSmoothness of(s) (p)i) Can be expressed as
Figure BDA0001150616040000081
The following two constraints are used to remove non-ground points:
A. removing points significantly above the road surface (z)i≥5m)(ziIs a point piHeight value of (d);
B. and removing the point of which the included angle between the normal vector and the Z axis is more than 22.5 degrees.
S13, calculating the hyper-voxel fi：
Thus, the point set P after the non-ground point is removedgAccording to the smoothness ranking of each point, the point with high smoothness is selected as a seed point. The hyper-voxels are computed in such a way that region growing starts from the seed point. Let voxel fiFormalized definition as a set of points P to which there belongsiCenter point oiAnd the normal vector nlFormed triplet
Figure BDA0001150616040000082
Seed for each seed pointiLet its initial hyper-voxel fiIs { p }iCenter point and normal vector are Tp (p) respectivelyi).oiAnd Tp (p)i).nl. Then, the principle of width priority is adopted to fiA region growing is performed. For each candidate point pjIf (1) p is satisfiedjTo piIs less than a threshold value Rseed(ii) a (2) Vector Tp (p)j).nlAnd Tp (p)i).nlThe included angle of the angle is less than 22.5 degrees; (3) p is a radical ofjTo Tp (p)i) Is less than a threshold e; then p will bejIs added to fiIs concentrated. When f isiWhen no longer expandable, according to fi.piFitting the plane using least squares, and fitting fi.nlg is updated as the normal vector of the fitted plane. And (3) assigning points to the hyper-voxels by adopting local K-means clustering on the basis of the initial facets, and ensuring that the distance from each point to the hyper-voxel to which the point belongs is smaller than the distance from each point to other hyper-voxels. The distance function here is defined as:
Figure BDA0001150616040000083
wherein Ds,DnAnd DiRespectively euclidean distance, normal vector distance and intensity distance. Omegas,ωnAnd ωiRespectively, the corresponding weights.
S2, extracting boundary points between adjacent non-coplanar hyper-voxels by using a α -shape algorithm;
after the point cloud is segmented into superpixels, for each superpixel, an α -shape algorithm can be used to extract boundary points, and meanwhile, the boundary points between two superpixels which are coplanar with each other are removed, namely, if the included angle of normal vectors of the two superpixels is less than 22.5 degrees, the boundary points between the two superpixels are deletedbIncluding road boundary points and non-road boundary points.
The α -shape algorithm can be viewed as an extension of the closure Convex Hull, which can compute a finer closure by adjusting the α parameters to describe roughly the shape of a group of points in a plane or space, specifically, a pair of points is nested with a circle of a certain fixed radius, and when a pair of points just falls on the circle and no other point is contained in the circle, the two points are the boundary points of the shape.
S3, extracting road boundary points by using an energy minimization algorithm based on graph cut;
road boundary points are next extracted using a graph cut based energy minimization algorithm. The vehicle-mounted laser scanning system provides vehicle running track line data, and the track line data is observed to be basically consistent with the position direction of a measured road. The trajectory data is thus used as an initial observation model for the graph cut algorithm. The energy formula is defined as:
E(f)＝Edata(f)+λ·Esmooth(f)
Figure BDA0001150616040000091
Figure BDA0001150616040000092
Figure BDA0001150616040000093
where P isbRefers to the set of boundary points extracted in step 2. n is piThe potential of the point set of the belonging superpixel. Δ djIs pointing at pjTo a straight line
Figure BDA0001150616040000094
The distance of (c). Δ diIs pointing at piAll points to straight lines in the neighborhood
Figure BDA0001150616040000095
Average redundancy of (2). Sigma1Mean redundancy of all points. Straight line
Figure BDA0001150616040000096
Is defined as passing through point piAnd the direction and the distance piThe direction of the nearest trajectory line.
Figure BDA0001150616040000097
Figure BDA0001150616040000098
Figure BDA0001150616040000101
Figure BDA0001150616040000102
(xi,yi,zi),(xj,yj,zj) Are respectively a point piAnd pjIs determined by the three-dimensional coordinates of (a),
Figure BDA0001150616040000103
is pointing at piAnd pjThe euclidean distance of (c). Here, the
Figure BDA0001150616040000104
Indicates if the point piAnd pjIf the allocated labels are consistent, the cost is zero, otherwise, the cost is
Figure BDA0001150616040000105
Where σ is2Refers to a set of points PbThe spatial resolution of (a). The result of using the graph cut algorithm to find the minimum value of the above energy formula is to divide the boundary points into two types, one is the road boundary points, and the other is the non-road boundary points.
S4, removing outliers based on an Euclidean distance clustering algorithm;
and clustering the obtained road boundary points by using an Euclidean distance clustering algorithm, and deleting the category with small point number, namely deleting the category if the number of the points contained in one category is less than 5 after clustering.
And S5, fitting the extracted road boundary points into a smooth curve.
The remaining classes are fitted to smooth curves, respectively, thereby obtaining the road boundaries. Here, Cubic Spline Interpolation (Cubic Spline Interpolation) is used to fit the line.
Fig. 3 is an effect diagram after the processing, showing the extracted road.
The above description is only for the preferred embodiment of the present invention, but the scope of the present invention is not limited thereto, and any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope of the present invention are included in the scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims (5)

1. A three-dimensional point cloud road boundary automatic extraction method is characterized by comprising the following steps: the method comprises the following steps:
s1, screening seed points of the obtained whole three-dimensional point cloud data set P for superpixel division;
s2, extracting boundary points between adjacent non-coplanar hyper-voxels by using a α -shape algorithm;
s3, extracting road boundary points by using an energy minimization algorithm based on graph cut;
s4, removing outliers based on an Euclidean distance clustering algorithm;
s5, fitting the road boundary points with the outliers removed into a smooth curve;
the process of dividing the hyper-voxels in step S1 includes the following steps:
s11 solving fitting plane Tp(Pi) Wherein P isiFor each input point of the entire three-dimensional point cloud dataset P;
s12, removing non-ground points;
s13, calculating the hyper-voxel fi；
And, solving the fitting plane Tp(Pi) The method comprises the following specific steps:
for each input point P of the entire three-dimensional point cloud dataset PiTangent plane T thereofp(Pi) Can be represented as a point o from its centeriSum normal vector nlThe binary set of components, namely:
Figure FDA0002349358200000011
wherein the content of the first and second substances,
Figure FDA0002349358200000012
is a normal vector nlIs determined by the estimated value of (c),
any point p to T in three-dimensional spacep(Pi) The distance of (d) can be expressed as:
Figure FDA0002349358200000013
note PiK neighbors of (a) are set of NbK(Pi) The best fit plane in the least squares sense can be obtained by solving the following equation:
Figure FDA0002349358200000014
the fitted plane is then optimized using an iterative reweighted least squares method:
Figure FDA0002349358200000021
obtaining the optimized fitting plane T by solving the weighted least square equationp(Pi)：
Figure FDA0002349358200000022
For the optimized fitting plane Tp(Pi) The above solving process is repeated until the algorithm for solving the weighted least squares equation converges.
2. The method for automatically extracting the road boundary by using the three-dimensional point cloud as claimed in claim 1, wherein the step S2 of extracting the boundary points between the adjacent non-coplanar hyper-voxels by using α -shape algorithm comprises the following specific steps:
after the point cloud is segmented into superpixels, for each superpixel, the α -shape algorithm may be used to extract boundary points while removing edges between two superpixels that are coplanar with each otherBoundary points, i.e. if the normal vector angle of two hyper-voxels is less than 22.5 °, the boundary point between the two hyper-voxels is deleted, the boundary point P being the point at this timebIncluding road boundary points and non-road boundary points.
3. The method for automatically extracting the road boundary by the three-dimensional point cloud according to claim 1, wherein the method comprises the following steps:
s12, removing the non-ground points, the concrete steps are as follows:
finally form fitting plane Tp(Pi) Three eigenvalues of the covariance matrix of the point set of (a) are λ1，λ2And λ3And satisfy lambda1≥λ2≥λ3Then point PiSmoothness of (S) (P)i) Can be expressed as:
Figure FDA0002349358200000023
the following two constraints are used to remove non-ground points:
A. removing points significantly above the road surface, i.e. ziA point of not less than 5m, ziIs a point PiA height value of (d);
B. and removing the point of which the included angle between the normal vector and the Z axis is more than 22.5 degrees.
4. The method for automatically extracting the road boundary by the three-dimensional point cloud according to claim 2, wherein the method comprises the following steps:
s13, calculating the hyper-voxel fiThe method comprises the following specific steps:
removing the point set P after the non-ground pointgSorting according to the smoothness of each point, firstly selecting the point with the smoothness larger than a preset value as a seed point, and calculating the hyper-voxels in a mode of starting region growing from the seed point; let voxel fiFormalized definition as a point PiCenter point oiAnd the normal vector nlFormed triplet
Figure FDA0002349358200000031
Seed for each seed pointiLet its initial hyper-voxel fiIs { P }iThe center point and normal vector are T respectivelyp(Pi).oiAnd Tp(Pi).nl(ii) a Then, the principle of width priority is adopted to fiPerforming region growing on each candidate point PjIf (1) P is satisfied at the same timejTo PiIs less than a threshold value Rseed(ii) a (2) Vector Tp(Pj).nlAnd Tp(Pi).nlThe included angle of the angle is less than 22.5 degrees; (3) pjTo Tp(Pi) Is less than a threshold e; then P will bejIs added to fiWhen f is concentratediWhen the expansion can not be performed any more, according to the triple fiP in (1)iThis term fi.PiFitting the plane using least squares and fitting the triplet fiIn (1)
Figure FDA0002349358200000032
This item
Figure FDA0002349358200000033
Updating the normal vector of the fitting plane; at all f calculatediOn the basis of the triple, points are assigned to the hyper-voxels by adopting local K-means clustering, and the distance from each point to the hyper-voxel to which the point belongs is ensured to be smaller than the distances to other hyper-voxels, wherein a distance function is defined as:
Figure FDA0002349358200000034
wherein Ds，DnAnd DiRespectively Euclidean distance, normal vector distance and intensity distance, omegas，ωnAnd ωiThe weights are corresponding to the three distance values, respectively.
5. The method for automatically extracting the three-dimensional point cloud road boundary according to claim 4, wherein the method comprises the following steps: step S3 extracts road boundary points using an energy minimization algorithm based on graph cut, as follows:
the method comprises the steps of taking vehicle running track line data provided by a vehicle-mounted laser scanning system as an initial observation model, dividing boundary points into the following two categories { 'road boundary points', 'non-road boundary points' } by using a graph cut algorithm, namely, solving a classification function f to assign a label f to each point by using the graph cut algorithmpE.g., L is a category set { "road boundary points", "non-road boundary points" }, so that the cost of payment is minimal, i.e., the energy formula is minimized,
the energy formula is defined here as:
E(f)＝Edata(f)+λ·Esmooth(f)
here Edata(f) I.e. the data item in the energy formula, refers to the error of the comparison of the classification result with the initial observation model, which is the cost of assigning a label to each point in the classification process, Esmooth(f) That is, the smooth term in the energy formula refers to the degree of non-smoothness of the classification function f, specifically, the cost of inconsistency of classification results between each point and neighboring points in the classification process, and λ is the smooth term Esmooth(f) Is set to 32, here empirically, wherein,
Figure FDA0002349358200000041
Figure FDA0002349358200000042
Figure FDA0002349358200000043
where P isbRefers to the set of boundary points extracted in step S2, n is PiPotential of point set of the supervoxel to which it belongs, Δ djIs pointing at PjTo a straight line
Figure FDA0002349358200000044
A distance of Δ diIs pointing at PiAll points to straight lines in the neighborhood
Figure FDA0002349358200000045
Average redundancy of, σ1Is a set of points PgMean redundant, straight line of all points in
Figure FDA00023493582000000414
Is defined as passing through point PiAnd the direction vector and the distance PiThe nearest trajectory line is in the same direction,
Figure FDA0002349358200000046
Figure FDA0002349358200000047
Figure FDA0002349358200000048
Figure FDA0002349358200000049
Figure FDA00023493582000000410
is PiThe label of (a) is used,
Figure FDA00023493582000000411
is Pj(ii) a label of (x)i，yi，zi)，(xj，yj，zj) Are respectively a point PiAnd PjIs determined by the three-dimensional coordinates of (a),
Figure FDA00023493582000000412
is pointing at PiAnd PjEuclidean distance ofFrom here, here
Figure FDA00023493582000000413
Indicates if the point P isiAnd PjIf the allocated labels are consistent, the cost is zero, otherwise, the cost is B { P }i，Pj}，
Where σ is2Refers to a set of points PbThe spatial resolution of (2) is obtained by using a graph cut algorithm to obtain the minimum value of the energy formula, namely, the boundary points are divided into two types, one type is road boundary points, and the other type is non-road boundary points.
CN201610996779.6A 2016-11-11 2016-11-11 Automatic extraction method for three-dimensional point cloud road boundary Active CN106780524B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610996779.6A CN106780524B (en) 2016-11-11 2016-11-11 Automatic extraction method for three-dimensional point cloud road boundary

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610996779.6A CN106780524B (en) 2016-11-11 2016-11-11 Automatic extraction method for three-dimensional point cloud road boundary

Publications (2)

Publication Number Publication Date
CN106780524A CN106780524A (en) 2017-05-31
CN106780524B true CN106780524B (en) 2020-03-06

Family

ID=58973201

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610996779.6A Active CN106780524B (en) 2016-11-11 2016-11-11 Automatic extraction method for three-dimensional point cloud road boundary

Country Status (1)

Country Link
CN (1) CN106780524B (en)

Families Citing this family (37)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107390152B (en) * 2017-07-14 2020-02-18 歌尔科技有限公司 Calibration method and device for magnetometer and electronic equipment
CN107516098B (en) * 2017-07-30 2021-08-10 华南理工大学 Target contour three-dimensional information extraction method based on edge curvature angle
CN107657659A (en) * 2017-08-14 2018-02-02 南京航空航天大学 The Manhattan construction method for automatic modeling of scanning three-dimensional point cloud is fitted based on cuboid
CN110044371A (en) * 2018-01-16 2019-07-23 华为技术有限公司 A kind of method and vehicle locating device of vehicle location
CN110068834B (en) * 2018-01-24 2023-04-07 北京京东尚科信息技术有限公司 Road edge detection method and device
CN108831146A (en) * 2018-04-27 2018-11-16 厦门维斯云景信息科技有限公司 Generate semi-automatic cloud method of three-dimensional high-definition mileage chart intersection lane
CN108615242B (en) * 2018-05-04 2021-07-27 重庆邮电大学 High-speed guardrail tracking method
CN109241978B (en) * 2018-08-23 2021-09-07 中科光绘（上海）科技有限公司 Method for rapidly extracting plane piece in foundation three-dimensional laser point cloud
CN109299739A (en) * 2018-09-26 2019-02-01 速度时空信息科技股份有限公司 The method that vehicle-mounted laser point cloud is filtered based on the surface fitting of normal vector
CN109635641B (en) * 2018-11-01 2020-06-09 百度在线网络技术（北京）有限公司 Method, device and equipment for determining road boundary line and storage medium
CN109584294B (en) * 2018-11-23 2020-08-28 武汉中海庭数据技术有限公司 Pavement point cloud extraction method and device based on laser point cloud
CN109815943B (en) * 2019-03-18 2021-02-09 北京石油化工学院 Hazardous chemical storage stacking picture sample generation method and system
CN110046661A (en) * 2019-04-10 2019-07-23 武汉大学 A kind of vehicle-mounted cloud clustering method cutting algorithm based on contextual feature and figure
CN110060338B (en) * 2019-04-25 2020-11-10 重庆大学 Prefabricated part point cloud identification method based on BIM model
CN110243369A (en) * 2019-04-30 2019-09-17 北京云迹科技有限公司 Drawing method and device are remotely built suitable for robot localization
CN110249741B (en) * 2019-06-05 2020-07-28 中国农业大学 Potato seed potato dicing method based on point cloud model
CN110222642B (en) * 2019-06-06 2021-07-16 上海黑塞智能科技有限公司 Plane building component point cloud contour extraction method based on global graph clustering
CN110349252B (en) * 2019-06-30 2020-12-08 华中科技大学 Method for constructing actual machining curve of small-curvature part based on point cloud boundary
CN110458083B (en) * 2019-08-05 2022-03-25 武汉中海庭数据技术有限公司 Lane line vectorization method, device and storage medium
CN110673107B (en) * 2019-08-09 2022-03-08 北京智行者科技有限公司 Road edge detection method and device based on multi-line laser radar
CN110967024A (en) * 2019-12-23 2020-04-07 苏州智加科技有限公司 Method, device, equipment and storage medium for detecting travelable area
CN111563457A (en) * 2019-12-31 2020-08-21 成都理工大学 Road scene segmentation method for unmanned automobile
CN111340935A (en) * 2020-01-23 2020-06-26 北京市商汤科技开发有限公司 Point cloud data processing method, intelligent driving method, related device and electronic equipment
CN111815776A (en) * 2020-02-04 2020-10-23 山东水利技师学院 Three-dimensional building fine geometric reconstruction method integrating airborne and vehicle-mounted three-dimensional laser point clouds and streetscape images
CN111487646A (en) * 2020-03-31 2020-08-04 安徽农业大学 Online detection method for corn plant morphology
CN111553946B (en) * 2020-04-17 2023-04-18 中联重科股份有限公司 Method and device for removing ground point cloud and method and device for detecting obstacle
CN111768421A (en) * 2020-07-03 2020-10-13 福州大学 Edge-aware semi-automatic point cloud target segmentation method
CN111783721B (en) * 2020-07-13 2021-07-20 湖北亿咖通科技有限公司 Lane line extraction method of laser point cloud and electronic equipment
CN112561808B (en) * 2020-11-27 2023-07-18 中央财经大学 Road boundary line restoration method based on vehicle-mounted laser point cloud and satellite image
CN112560747A (en) * 2020-12-23 2021-03-26 苏州工业园区测绘地理信息有限公司 Vehicle-mounted point cloud data-based lane boundary interactive extraction method
CN112731338A (en) * 2020-12-30 2021-04-30 潍柴动力股份有限公司 Storage logistics AGV trolley obstacle detection method, device, equipment and medium
CN112733696B (en) * 2021-01-04 2023-08-15 长安大学 Vehicle-mounted LIDAR road edge extraction method based on multi-model fitting
CN113409332B (en) * 2021-06-11 2022-05-27 电子科技大学 Building plane segmentation method based on three-dimensional point cloud
CN113378800B (en) * 2021-07-27 2021-11-09 武汉市测绘研究院 Automatic classification and vectorization method for road sign lines based on vehicle-mounted three-dimensional point cloud
CN114972377A (en) * 2022-05-24 2022-08-30 厦门大学 3D point cloud segmentation method and device based on moving least square method and hyper-voxels
CN117576144B (en) * 2024-01-15 2024-03-29 湖北工业大学 Laser point cloud power line extraction method and device and electronic equipment
CN117726775A (en) * 2024-02-07 2024-03-19 法奥意威(苏州)机器人系统有限公司 Point cloud preprocessing method and device based on grid downsampling

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103559737A (en) * 2013-11-12 2014-02-05 中国科学院自动化研究所 Object panorama modeling method

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
GB2532948B (en) * 2014-12-02 2021-04-14 Vivo Mobile Communication Co Ltd Object Recognition in a 3D scene

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN103559737A (en) * 2013-11-12 2014-02-05 中国科学院自动化研究所 Object panorama modeling method

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
3D ROAD SURFACE EXTRACTION FROM MOBILE LASER SCANNING POINT CLOUDS;Dawei Zai et al;《2016 IEEE international geoscience and remote sensing symposium》;20160715;第1595-1598页 *
A parallel point cloud clustering algorithm for subset segmentation and outlier detection;Christian Teutsch et al;《SPIE Optical Metrology》;20111231;第8085卷;第1-6 *
Robust Piecewise-Planar 3D Reconstruction and Completion from Large-Scale Unstructured Point Data;Anne-Laure Chauve et al;《2010 IEEE Computer Society Conference on Computer Vision and Pattern Recognition》;20100618;第1261-1268页 *

Also Published As

Publication number Publication date
CN106780524A (en) 2017-05-31

Similar Documents

Publication Publication Date Title
CN106780524B (en) Automatic extraction method for three-dimensional point cloud road boundary
Vosselman et al. Recognising structure in laser scanner point clouds
Zolanvari et al. Three-dimensional building façade segmentation and opening area detection from point clouds
Lee et al. Fusion of lidar and imagery for reliable building extraction
Cheng et al. Integration of LiDAR data and optical multi-view images for 3D reconstruction of building roofs
Matei et al. Building segmentation for densely built urban regions using aerial lidar data
Kedzierski et al. Methods of laser scanning point clouds integration in precise 3D building modelling
CN111652241B (en) Building contour extraction method integrating image features and densely matched point cloud features
CN115564926B (en) Three-dimensional patch model construction method based on image building structure learning
Chen et al. Building reconstruction from LIDAR data and aerial imagery
KR101549155B1 (en) Method of automatic extraction of building boundary from lidar data
CN116449384A (en) Radar inertial tight coupling positioning mapping method based on solid-state laser radar
CN112669333A (en) Single tree information extraction method
CN113409332B (en) Building plane segmentation method based on three-dimensional point cloud
JP2023530449A (en) Systems and methods for air and ground alignment
Omidalizarandi et al. Segmentation and classification of point clouds from dense aerial image matching
Sun et al. Automated segmentation of LiDAR point clouds for building rooftop extraction
Zeybek et al. Road surface and inventory extraction from mobile LiDAR point cloud using iterative piecewise linear model
CN112330604A (en) Method for generating vectorized road model from point cloud data
Ni et al. Applications of 3d-edge detection for als point cloud
JP2003141567A (en) Three-dimensional city model generating device and method of generating three-dimensional city model
CN117253205A (en) Road surface point cloud rapid extraction method based on mobile measurement system
Novacheva Building roof reconstruction from LiDAR data and aerial images through plane extraction and colour edge detection
Mi et al. Automatic road structure detection and vectorization Using Mls point clouds
CN107808160B (en) Three-dimensional building extraction method and device

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
TR01 Transfer of patent right
TR01 Transfer of patent right

Effective date of registration: 20231214

Address after: 361000 Siming South Road, Xiamen, Fujian Province, No. 422

Patentee after: XIAMEN University

Patentee after: National University of Defense Technology

Address before: 361000 Siming South Road, Xiamen, Fujian Province, No. 422

Patentee before: XIAMEN University