CN110443836B - Point cloud data automatic registration method and device based on plane features - Google Patents
Point cloud data automatic registration method and device based on plane features Download PDFInfo
- Publication number
- CN110443836B CN110443836B CN201910550064.1A CN201910550064A CN110443836B CN 110443836 B CN110443836 B CN 110443836B CN 201910550064 A CN201910550064 A CN 201910550064A CN 110443836 B CN110443836 B CN 110443836B
- Authority
- CN
- China
- Prior art keywords
- plane
- point cloud
- cloud data
- registered
- triple
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20021—Dividing image into blocks, subimages or windows
Abstract
The invention relates to a point cloud data automatic registration method and device based on plane features, and belongs to the technical field of three-dimensional laser scanning. The method comprises the steps of firstly carrying out plane segmentation on point cloud data through a plane segmentation algorithm to obtain plane pieces, calculating attribute information of each plane, then establishing a corresponding relation among the plane pieces through the attribute information of the plane pieces, the mutual relation among the planes and a rotation and translation geometric constraint to obtain a corresponding plane pair set for solving coordinate conversion parameters, solving the coordinate conversion parameters through the plane parameters, and selecting an optimal solution, namely a final registration result, according to the established point cloud registration overall consistency measurement. The method comprehensively utilizes the plane attribute information, the plane constraint and the space geometric constraint to ensure the accuracy and the high efficiency of the registration process, does not depend on additional strength or color information, can effectively register point cloud data acquired by various platforms, and has strong applicability.
Description
Technical Field
The invention relates to a point cloud data automatic registration method and device based on plane features, and belongs to the technical field of three-dimensional laser scanning.
Background
The three-dimensional point cloud is a discrete sampling of an objective world, is composed of a series of points, and can be obtained by a laser scanner, a laser radar (LiDAR), a depth camera, a stereoscopic vision and other sensors. In the field of surveying and mapping remote sensing, three-dimensional laser scanning is known as a 'live-action replication' technology, and mass point cloud data of a target can be quickly and accurately acquired, so that great attention is paid to the technology. Due to the limited effective range and view field range of the measuring instrument and the influence of shielding, in order to obtain the complete point cloud of the operation scene, the measuring instrument is often erected at a plurality of measuring stations for measurement, and then sequence registration is performed among the point clouds, as shown in fig. 1. The essence of point cloud registration is to seek a set of translation and rotation parameters to unify the point clouds in the same coordinate system, so that the overlapped parts of the point clouds after coordinate transformation are well aligned. Describing a point cloud registration problem by a mathematical language, and giving a point cloud P to be registered＝{pi＝(xi,yi,zi) Q and the target point cloud Q ═ Qi＝(xi,yi,zi)}，
Solving rigid body transformations
Such that:
wherein R ∈ SO (3) is a rotation matrix, t is a translation vector, eijFor error measurement, C { (i, j)mDenotes a corresponding point pair set, point piCorresponding point qj。
In the last two decades, three-dimensional laser scanning has become a standard technology in many fields by virtue of its specific advantages, is extensively researched and widely applied, and particularly, with the continuous improvement of cost performance, various products are continuously updated and updated, and the three-dimensional laser scanning has been applied to the fields of three-dimensional modeling, digital factories, cultural relic protection, landslide monitoring, asset management and the like. However, the massive amount of point cloud obtained by the three-dimensional laser scanner poses a challenge to the later data processing. In general measurement operation, registration among multi-station point clouds is often realized by laying artificial targets, and by adopting the method, a target laying scheme needs to be designed in advance, secondary fine scanning needs to be carried out on the targets in the measurement process, time and labor are wasted, and extra cost needs to be paid for target purchase. In addition, in many cases (such as tall and big city building groups), the targets manually laid can only be distributed in a small space range, and good mesh constraint cannot be formed on the registration parameters, so that the coordinate conversion parameters obtained by solving have large errors; even in some cases where it is not possible or convenient to lay out the target, the registration process of the point cloud will be very annoying to the field handler. When point cloud registration can not be realized based on a manual target, more than three groups of corresponding points are often selected from the point cloud to be registered manually to calculate a rough initial registration result, and the rough initial registration result is used as an initial valueThe final registration result is solved by using an Iterative Closest Point (ICP), which consumes a lot of energy and time and cannot meet the requirement of Point cloud data automation processing. Meanwhile, the massive amount of the point clouds puts higher requirements on a data processing hardware platform, three-dimensional coordinate points are directly used as processing elements, when dozens of or even hundreds of groups of point clouds are processed, a large memory is occupied, and a Central Processing Unit (CPU) and a Graphic Processing Unit (GPU) with ultrahigh performance are needed. The search for automatic and efficient point cloud registration has been a hot research problem in the field of point cloud data processing,
the method comprises the steps of utilizing intensity information to assist point cloud registration, firstly projecting three-dimensional point cloud into a two-dimensional intensity image, extracting feature points through the intensity image and calculating a descriptor so as to establish a corresponding relation, and solving coordinate conversion parameters. However, there is inevitable distortion in projecting the point cloud into an image, and the intensity value is greatly affected by the scanning distance and angle, and the accuracy of point cloud registration is greatly affected.
In an actual measurement scene, a large number of geometric features such as angular points, planes, straight line segments, cylinders and the like often exist, the point cloud registration can be achieved by reasonably utilizing the features, the extraction of the plane features is relatively easy to achieve, and the point cloud registration is widely distributed in an artificial scene, so that the point cloud registration based on the extracted plane features is a potential feasible method. The existing point cloud registration method based on plane features is mainly divided into the following two types: (1) an additional auxiliary device is needed, such as a chinese patent application document with application publication No. CN105571519A, which discloses an auxiliary device for point cloud stitching of a three-dimensional scanner and a stitching method thereof, the method also needs to perform scanning measurement operation according to a predetermined requirement in order to realize registration, so that data acquisition is not flexible enough, and the method is not suitable for large scenes; (2) for example, the chinese patent application publication No. CN106570823A discloses a point cloud rough stitching method based on matching of plane features, which further needs to manually select and determine corresponding planes in subsequent processing, that is, manually perform plane matching, but cannot realize automation through a program, and manually select corresponding planes to realize registration, which often only can select a few pairs of corresponding planes from a large number of plane features, and is difficult to select an optimal plane combination to participate in registration resolution, so that only a rough registration result with a large error can be calculated. In addition, when matching the corresponding planes, the existing registration method based on the plane features is usually based on a complete plane hypothesis, that is, the corresponding planes in the two groups of point clouds are required to be completely consistent, so that the requirements on the overlapping degree between the target point cloud and the point cloud to be registered and the environmental shielding are high, and when the overlapping degree between the point clouds is low or the shielding is serious, the point cloud registration is difficult to realize. Therefore, the current point cloud registration method based on the plane features has limited use scenes, the error of the registration result is large, and the automation degree needs to be improved.
Disclosure of Invention
The invention aims to provide a point cloud data automatic registration method and device based on plane features, and aims to solve the problems of large error, poor adaptability and low automation degree of the existing point cloud registration method.
The invention provides a point cloud data automatic registration method based on plane features for solving the technical problems, which comprises the following steps:
1) acquiring target point cloud data and point cloud data to be registered;
2) respectively carrying out plane segmentation on the target point cloud data and the point cloud data to be registered to obtain a target plane piece group and a plane piece group to be registered, and calculating attribute information of each plane piece in each plane group;
3) obtaining a plane pair set with consistent attributes from a target plane piece set and a plane piece set to be registered according to the attribute information of each plane piece;
4) selecting two pairs of non-parallel plane pairs from the plane pair sets with consistent attributes, and determining a rotation matrix according to the obtained two pairs of non-parallel plane pairs; selecting a non-parallel plane pair meeting rotation consistency from the remaining plane pairs of the plane pair set with consistent attributes by using a rotation matrix, taking the three selected plane pairs as a triple, acting conversion parameters corresponding to the triple on a plane of point cloud to be registered in the triple, and adding the triple into the triple set if an overlapping condition is met;
5) traversing the remaining plane pairs in the plane pair set with consistent attributes according to the mode of the step 4) to obtain a triple set, respectively acting the conversion parameters corresponding to each triple in the triple set on the cloud data of the point to be registered, calculating the overall consistency measurement, and selecting the conversion parameter of the triple with the minimum overall consistency measurement from the triple to perform point cloud registration.
The invention also provides a plane feature-based point cloud data automatic registration device, which comprises a memory, a processor and a computer program, wherein the computer program is stored on the memory and runs on the processor, the processor is coupled with the memory, and the processor realizes the plane feature-based point cloud data automatic registration method when executing the computer program.
The method comprises the steps of firstly carrying out plane segmentation on point cloud data through a plane segmentation algorithm to obtain plane pieces, calculating attribute information of each plane, then establishing a corresponding relation among the plane pieces through the attribute information of the plane pieces, the mutual relation among the planes and a rotation and translation geometric constraint to obtain a corresponding plane pair set for solving coordinate conversion parameters, solving the coordinate conversion parameters for each group of plane pairs, and selecting an optimal solution, namely a final registration result, according to the established point cloud registration overall consistency measurement. The method comprehensively utilizes the plane attribute information, the plane constraint and the space geometric constraint to ensure the accuracy and the high efficiency of the registration process, does not depend on additional strength or color information, can effectively register point cloud data acquired by various platforms, has strong applicability, and can process small-scene indoor point clouds and large-scale complex outdoor scene point clouds.
Further, in order to avoid unnecessary interference and reduce the complexity of subsequent processing, the method also comprises the step of preprocessing the target point cloud data and the to-be-registered point cloud data before plane segmentation, and removing in-vitro isolated points and outlier points from the target point cloud data and the to-be-registered point cloud data.
Further, the attribute information of each planar patch in step 2) includes a unit normal vector of the plane, a distance from the coordinate origin to the fitting plane, a mean square error of the plane fitting, an area of the planar patch, a boundary point of the planar patch, a boundary length, and a length and a width of a minimum bounding rectangle of the planar patch.
Further, the invention also provides a specific judgment mode of consistent attributes, wherein the plane pair set with consistent attributes in the step 3) refers to a plane pair which satisfies the condition that the plane area ratio is smaller than a set ratio, the difference between the plane shapes is smaller than a set shape difference and the difference between the plane fitting mean square errors is smaller than a set error in the target plane sheet set and the plane sheet set to be aligned.
Further, in order to improve the efficiency and the precision of plane extraction, the plane segmentation in the step 2) is realized by adopting an adaptive plane segmentation algorithm based on voxel growth.
Further, in order to improve the extraction efficiency, in the step 2), an incremental PCA algorithm is adopted to calculate the plane parameters after the growth of the voxels and the mean square error of plane fitting in the voxel growth process.
Further, to avoid the incremental PCA being susceptible to noise and outliers, the method further includes recalculating the planar parameters obtained using the incremental PCA algorithm by a plane fitting algorithm that accounts for the noise of the sensor measurements.
Further, in order to improve the accuracy of the rotation matrix calculation, the rotation matrix in the step 4) is calculated by using a singular value decomposition method of covariance matrix weighting.
Further, in order to improve the efficiency and robustness of matching, the method further comprises filtering the planar sheet in step 2) to remove the planar sheet with a small area, approximate line shape and poor flatness.
Drawings
FIG. 1 is a schematic diagram of point cloud registration;
FIG. 2 is a flow chart of the automatic registration method of point cloud data based on plane features according to the present invention.
Detailed Description
The following further describes embodiments of the present invention with reference to the drawings.
Method embodiment
The automatic point cloud data registration method based on the plane features comprises the steps of firstly, carrying out plane segmentation on target point cloud data and point cloud data to be registered to obtain a target plane piece group and a plane piece group to be registered, and calculating attribute information of each plane piece in each plane group; then according to the attribute information of each plane slice, obtaining a plane pair set with consistent attributes from the target plane slice set and the plane slice set to be registered; obtaining a corresponding plane pair set for solving coordinate conversion parameters according to the plane piece attribute information, the inter-plane correlation and the plane pair set with consistent rotation and translation geometric constraint dependency; and finally, selecting an optimal solution from the corresponding plane pair set as a final registration result according to the established point cloud registration overall consistency measurement. The implementation process of the method is shown in fig. 2, and the specific implementation steps are as follows.
1. And acquiring target point cloud data and point cloud data to be registered.
In this embodiment, the target point cloud data and the cloud data of the points to be registered are obtained through the three-dimensional laser scanner, and in order to avoid interference of some isolated points and outliers, the target point cloud data and the cloud data of the points to be registered obtained through the three-dimensional laser scanner need to be preprocessed, points far away from the origin of coordinates are removed, points within 80% of the effective measurement range of the three-dimensional laser scanner are reserved, and external isolated points and outliers are removed.
2. And performing plane segmentation on the target point cloud data and the point cloud data to be registered to obtain a target plane piece group and a plane piece group to be registered.
The plane segmentation can be realized by methods such as dynamic clustering, hough transform, Random Sample Consensus (RANSAC), voxel growth adaptive plane segmentation algorithm, and the like. In the embodiment, a self-adaptive plane segmentation algorithm based on voxel growth is adopted to automatically extract plane features in target point cloud data and point cloud data to be registered, and in order to improve efficiency in the voxel growth process, an incremental pca (principal component analysis) algorithm is adopted to calculate plane parameters after voxel growth and mean square error (a measure of flatness) of plane fitting. The specific process is as follows:
adopting a Hessian standard type to express that a plane equation is n & r ═ d, wherein r is the coordinate of a certain point on a plane, n is a unit normal vector of the plane, d is the distance from a coordinate origin to a fitting plane, and the incremental PCA plane fitting algorithm is described as follows:
two sets of point sets P1And P2Each of which includes n1And n2Point of which is indicated by riRepresenting point coordinates, m representing the center of gravity;
order matrix
The geometric information after the two sets of point sets are combined can be obtained by the following formula:
A＝A1+A2
wherein the content of the first and second substances,
is the eigenvector, σ, corresponding to the smallest eigenvalue of the matrix S2The mean square error of the plane fit.
3. Plane parameters are calculated for each planar patch in each plane group.
And calculating plane parameters of the series of plane slices obtained in the last step, wherein the plane parameters mainly refer to the distances from unit normal vectors and coordinate origins of the planes to the fitting plane. Because the incremental PCA adopted in the plane segmentation process is easily influenced by noise and abnormal points and the plane parameter calculation is not accurate enough, the plane parameter is recalculated by adopting a plane fitting algorithm considering the noise measured by the sensor, and the points are expressed as a homogeneous coordinate form x ═ rT,1]Let pi be [ n ]T,d]，
The superscript e represents unitization, and at this time, the plane equation can be represented as pi · x — 0, and the specific calculation steps are as follows:
(1) computing matrices
j ═ 1.. times, N, initialize c ═ 0, let
wj1 where ρjDistance of a point from the origin of coordinates, CjIs a covariance matrix, wjIs a weight factor;
(2) calculating the matrices M and N using the following equations, respectively
Computing matrices
And its corresponding unit eigenvector
If λ ≈ 0, return
c and
otherwise, c is updated according to the following formula,
and wj：
And (5) returning to the step (2) and carrying out iterative solution.
Finally, the plane parameters are found using the following formula:
order to
Wherein I is an identity matrix, and the covariance matrix of the plane parameters is:
in addition, the calculation of the attribute information includes the area of the planar patch, the boundary point of the planar patch, the boundary length, and the length and width of the rectangle surrounded by the planar patch. The general plane point set area and boundary point calculation method takes longer time, and the method accelerates the process by adopting a method based on a point cloud projection image. Firstly, the average point density is used as the projection resolution, and the average point density is used as the projection resolutionThe point set is projected to be a binary image, then the projected image is processed by mathematical morphology, the boundary pixel can be easily identified by using an image boundary detection algorithm, the boundary point of the plane sheet can be determined according to the mapping relation between the pixel and the point, the three-dimensional boundary point can be converted into two-dimensional according to the first and second main directions of the plane in the projection process, thereby being convenient for calculating the plane attribute information and obtaining a series of two-dimensional boundary point coordinates { p) projected on the planei＝(xi,yi),|i＝1,…,Nb}，NbFor the number of boundary points, the length L and width W of the bounding rectangle can be statistically determined from the coordinate values, and the plane area s and the boundary length L can be calculated by the following equations.
4. Filtering with a flat sheet.
In order to improve the efficiency of subsequent plane matching and the robustness of the algorithm, the plane slice filtering is required to be carried out, the plane slice with a smaller area is removed according to a preset parameter threshold, the plane slice with approximate linearity is removed according to the length-width ratio, and the plane slice with poor flatness, namely a larger plane fitting mean square error is removed.
5. The planar pieces are automatically matched.
Through the steps 1-4, the target point cloud data and the plane pieces corresponding to the point cloud data to be registered can be obtained respectively, wherein the target point cloud data obtains m plane pieces, and the point cloud data to be registered obtains n plane pieces which are called a target plane piece group and a plane piece group to be registered respectively.
(1) A set of attribute-consistent pairs of planes is determined.
According to the plane attribute, traversing the two plane sheet groups to obtain KacFor plane pair with consistent attributes
The superscripts t and s are used for marking the target point cloud and the point cloud to be registered, P represents a plane, and the attribute consistency refers to the following attribute measurement.
Ratio of plane area
Difference in shape
Difference of mean square error of plane fitting
The invention adopts relative quantity rather than absolute quantity for condition judgment, which is convenient for parameter setting, improves the registration success rate and enhances the robustness of the registration method to the overlapping degree change and the shielding between the point clouds. That is, if the attribute difference between the two planes is within a certain range, the algorithm will determine that the plane attributes are consistent, and store the consistent plane attributes into the consistent plane attribute set Θac。
(2) A rotation matrix is determined.
The rotation matrix only requires two pairs of non-parallel planes, thus traversing the set Θ of plane-attribute-consistent plane pairsacTwo pairs of non-parallel planar pairs are sought, the non-parallel planar pairs being pairs of planes that satisfy the following condition.
Condition 1:
condition 2:
condition 3:
wherein two groups of corresponding levels selected in Condition 1The faces need to satisfy a non-parallel relationship,
is a non-parallel angle threshold, the condition 2 indicates that two groups of planes need to meet the consistency of included angles, namely, the included angles of normal vectors of two planes of the target point cloud and the included angles of normal vectors of two planes of the point cloud to be registered need to be consistent,
is the corresponding angle difference threshold, condition 3 denotes the plane center of gravity (denoted r)cRepresenting) distance change needs to meet certain threshold conditions, thereby reducing the possibility of mismatching and greatly reducing the processing burden of subsequent algorithms. After two groups of corresponding planes meeting the three conditions are obtained, calculating a rotation matrix from a plane normal vector by a two-step rotation method:
(3) a triplet is determined.
Continuously traversing theta on the basis of the step (2)acIn the middle residual plane pair, searching the third group of corresponding planes meeting the conditions
To determine the translation parameters, the set of planes need to satisfy the rotational consistency, i.e. by passing through
After rotation matrix transformation, to be registered in the point cloud
Should be in accordance with
Parallel, i.e. satisfy
Where ψ is an angle threshold. The three sets of plane pairs constitute a plane-to-triplet for calculating translation parameters:
and simultaneously updating the rotation parameters:
converting parameters
And
and acting on a plane of the point cloud to be registered in the triple, performing overlapping judgment by using plane boundary points and the gravity center, if the overlapping condition is met, keeping the triple as a group of possible solutions, and if not, giving up. After this step N is obtainedΠGroup triplet
(4) And determining the triple set and the corresponding coordinate conversion parameters.
Traversing the obtained triple set, and expanding each triple Π, namely judging the theta from the theta through rotation and translation consistency and overlappingacThe plane pair meeting the condition is selected to be added into pi, and the expanded triple is represented by gamma. For each Γ, a new coordinate transformation parameter may be calculated, typically using weighted least squares, where a rotation matrix (also called a rotation parameter) may be calculated by:
wherein, the matrix W is a weight matrix, and the matrix N is obtained by accumulating corresponding unit normal vectors
This direct solution, however, may result in inaccurate results, given the presence of measurement noise. The solution is carried out by using a singular value decomposition method of covariance matrix weighting, the obtained rotation matrix is ensured to be a unit orthogonal matrix, and a correlation matrix Q is defined assNTWtN, wherein the weight is the reciprocal of the trace of the normal vector N fitting covariance matrix, and Q ═ UΛ V can be obtained by performing singular value decomposition on the traceTThe rotation matrix to be solved is:
order to
The translation vector (translation parameter) is calculated by the following formula, wherein the weight is determined by the inverse of the variance of the parameter d.
But here updated conditionally after new rotation and translation parameters are calculated.
For the
And
and respectively calculating rotation consistency measures according to the plane corresponding relation provided by the gamma and the following formula:
if it is
Accepting the rotation update, otherwise refusing the update, and maintaining the original value
For translation
And
similarly, a translation consistency metric is calculated according to:
if it is
Accepting translation updating, or refusing updating, and maintaining original value
(5) And selecting an optimal solution.
After the step (4) is finished, a series of feasible solutions are obtained, an optimal solution is required to be selected from the feasible solutions, for each feasible solution, the feasible solution acts on a plane to be registered, the corresponding relation is determined again by using a nearest neighbor rule, and the total consistency measure is calculated as follows:
and selecting the solution with the minimum delta value as a final registration result.
Device embodiment
The automatic registration device for point cloud data of plane features comprises a memory, a processor and a computer program which is stored on the memory and runs on the processor, wherein the processor is coupled with the memory, and the processor executes the computer program to realize the automatic registration method for point cloud data of plane features in the method embodiment.
In the plane segmentation process, the traditional point-based method is replaced by a voxel growth-based mode, and plane parameters are updated by combining an incremental PCA algorithm, so that the efficiency of extracting plane features from large-scale point clouds is greatly improved; filtering out partial non-ideal planes, limiting the number of the planes sent into the plane matching module within a certain range, and reducing the complexity of plane matching; in the plane matching process, the high efficiency of the registration process is further ensured by comprehensively utilizing the plane attribute information, the inter-plane constraint and the space geometric constraint; the whole process realizes automatic processing without manual intervention. Meanwhile, the invention only utilizes three-dimensional point coordinates, namely geometric information, and does not depend on additional intensity or color information, can effectively register point cloud data acquired by various platforms, and can process small-scene indoor point clouds and large-scale complex outdoor scene point clouds.
Claims (9)
1. A point cloud data automatic registration method based on plane features is characterized by comprising the following steps:
1) acquiring target point cloud data and point cloud data to be registered;
2) respectively carrying out plane segmentation on the target point cloud data and the point cloud data to be registered to obtain a target plane piece group and a plane piece group to be registered, and calculating attribute information of each plane piece in each plane group;
3) obtaining a plane pair set with consistent attributes from a target plane piece set and a plane piece set to be registered according to the attribute information of each plane piece; the plane pair set with consistent attributes refers to a plane pair which satisfies the condition that the area ratio of planes in the target plane sheet set and the plane sheet set to be aligned is smaller than a set ratio, the difference of the plane shapes is smaller than a set shape difference, and the difference of the plane fitting mean square errors is smaller than a set error;
4) selecting two pairs of non-parallel plane pairs from the plane pair sets with consistent attributes, and determining a rotation matrix according to the obtained two pairs of non-parallel plane pairs; selecting a non-parallel plane pair meeting rotation consistency from the remaining plane pairs of the plane pair set with consistent attributes by using a rotation matrix, taking the three selected plane pairs as a triple, acting conversion parameters corresponding to the triple on a plane of point cloud to be registered in the triple, and adding the triple into the triple set if an overlapping condition is met;
the non-parallel plane pair refers to a plane pair satisfying the following three conditions:
condition 1:
condition 2:
condition 3:
wherein the content of the first and second substances,
is the normal vector of the two planes to which the target point cloud belongs,
is two plane normal vectors to which the point cloud to be registered belongs,
is a non-parallel angle threshold value that is,
is the corresponding angle difference threshold;
respectively correspond to and represent
The center of gravity of the plane is located, and gamma represents a set threshold;
rotational consistency metric δRThe calculation formula of (2) is as follows:
the superscripts t and s are used for marking a target point cloud and a point cloud to be registered, and n is a unit normal vector of a plane;
the overall consistency measure δ is calculated as:
wherein m represents the center of gravity, and d is the distance from the coordinate origin to the fitting plane;
5) traversing the remaining plane pairs in the plane pair set with consistent attributes according to the mode of the step 4) to obtain a triple set, respectively acting the conversion parameters corresponding to each triple in the triple set on the cloud data of the point to be registered, calculating the overall consistency measurement, and selecting the conversion parameter of the triple with the minimum overall consistency measurement from the triple to perform point cloud registration.
2. The method of claim 1, further comprising the step of pre-processing the target point cloud data and the point cloud data to be registered before performing the planar segmentation, and removing isolated points and outliers from the target point cloud data and the point cloud data to be registered.
3. The method of claim 1, wherein the attribute information of each plane piece in the step 2) comprises a unit normal vector of the plane, a distance from an origin of coordinates to a fitting plane, a mean square error of plane fitting, an area of the plane piece, a boundary point of the plane piece, a boundary length, and a length and a width of a minimum bounding rectangle of the plane piece.
4. The method for automatically registering point cloud data based on planar features according to claim 3, wherein the planar segmentation in the step 2) is realized by adopting an adaptive planar segmentation algorithm based on voxel growth.
5. The method for automatically registering point cloud data based on plane features according to claim 4, wherein the step 2) adopts an incremental PCA algorithm to calculate plane parameters after voxel growth and mean square error of plane fitting in the voxel growth process.
6. The method of claim 5, further comprising recalculating the planar parameters from the incremental PCA algorithm by a plane fitting algorithm that accounts for sensor measurement noise.
7. The method for automatically registering point cloud data based on plane features as claimed in claim 1, wherein the rotation matrix in the step 4) is calculated by using a singular value decomposition method of covariance matrix weighting.
8. The method for automatically registering point cloud data based on planar features of any one of claims 1-7, wherein the method further comprises filtering the planar patches in step 2) to remove planar patches with small area, approximate line shape and poor flatness.
9. An apparatus for automatic registration of point cloud data based on planar features, the apparatus comprising a memory and a processor, and a computer program stored on the memory and running on the processor, the processor being coupled to the memory, the processor implementing the method for automatic registration of point cloud data of planar features according to any one of claims 1 to 8 when executing the computer program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910550064.1A CN110443836B (en) | 2019-06-24 | 2019-06-24 | Point cloud data automatic registration method and device based on plane features |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910550064.1A CN110443836B (en) | 2019-06-24 | 2019-06-24 | Point cloud data automatic registration method and device based on plane features |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110443836A CN110443836A (en) | 2019-11-12 |
| CN110443836B true CN110443836B (en) | 2022-03-18 |
Family
ID=68428272
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910550064.1A Active CN110443836B (en) | 2019-06-24 | 2019-06-24 | Point cloud data automatic registration method and device based on plane features |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110443836B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11787051B2 (en) * | 2021-11-25 | 2023-10-17 | Nanjing University Of Aeronautics And Astronautics | Method for automatically processing structure-reinforcing member of aircraft |
Families Citing this family (26)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111161327A (en) * | 2019-12-31 | 2020-05-15 | 电子科技大学 | Point cloud registration method combining rotary platform and ICP (inductively coupled plasma) |
| CN111325779B (en) * | 2020-02-07 | 2020-12-11 | 贝壳找房(北京)科技有限公司 | Point cloud registration method and device, electronic equipment and storage medium |
| CN111582285B (en) * | 2020-04-28 | 2023-05-23 | 清华大学 | Plane extraction method of point cloud image |
| CN111768490B (en) * | 2020-05-14 | 2023-06-27 | 华南农业大学 | Plant three-dimensional modeling method and system based on iteration closest point and manual intervention |
| CN111899291A (en) * | 2020-08-05 | 2020-11-06 | 深圳市数字城市工程研究中心 | Automatic registration method for coarse-to-fine urban point cloud based on multi-source dimension decomposition |
| CN112099039A (en) * | 2020-08-25 | 2020-12-18 | 西北工业大学 | Indoor relative positioning method based on laser radar |
| CN112132875B (en) * | 2020-08-31 | 2023-07-28 | 青岛秀山移动测量有限公司 | Multi-platform point cloud matching method based on surface features |
| CN112268559B (en) * | 2020-10-22 | 2023-03-28 | 中国人民解放军战略支援部队信息工程大学 | Mobile measurement method for fusing SLAM technology in complex environment |
| CN112232248B (en) * | 2020-10-22 | 2023-04-07 | 中国人民解放军战略支援部队信息工程大学 | Method and device for extracting plane features of multi-line LiDAR point cloud data |
| CN114526720B (en) * | 2020-11-02 | 2024-04-16 | 北京四维图新科技股份有限公司 | Positioning processing method, device, equipment and storage medium |
| CN112767458B (en) * | 2020-11-13 | 2022-07-29 | 武汉中海庭数据技术有限公司 | Method and system for registering laser point cloud and image |
| CN112489099B (en) * | 2020-12-11 | 2023-01-20 | 北京航空航天大学 | Point cloud registration method and device, storage medium and electronic equipment |
| CN112884902B (en) * | 2021-03-17 | 2023-03-17 | 中山大学 | Point cloud registration-oriented target ball position optimization method |
| CN113052881A (en) * | 2021-03-24 | 2021-06-29 | 王程 | Automatic registration method for extracting pole point in indoor three-dimensional point cloud |
| CN113223067B (en) * | 2021-05-08 | 2021-11-19 | 广东三姆森科技股份有限公司 | Online registration method for incomplete three-dimensional scanning point cloud with plane reference |
| CN113516695B (en) * | 2021-05-25 | 2023-08-08 | 中国计量大学 | Point cloud registration strategy in laser profiler flatness measurement |
| CN113344992B (en) * | 2021-05-31 | 2022-06-28 | 山东大学 | Global point cloud registration method, system, storage medium and equipment |
| CN114241018B (en) * | 2021-12-09 | 2023-05-09 | 电子科技大学 | Tooth point cloud registration method, system and readable storage medium |
| CN114463396B (en) * | 2022-01-07 | 2024-02-06 | 武汉大学 | Point cloud registration method utilizing plane shape and topological graph voting |
| CN115100254B (en) * | 2022-06-10 | 2023-06-30 | 兰州交通大学 | Point cloud registration method for dual quaternion description under plane feature constraint |
| CN115439644B (en) * | 2022-08-19 | 2023-08-08 | 广东领慧数字空间科技有限公司 | Similar point cloud data alignment method |
| CN116883469B (en) * | 2023-07-20 | 2024-01-19 | 中国矿业大学 | Point cloud registration method based on EIV model description under plane feature constraint |
| CN117173227B (en) * | 2023-11-01 | 2024-01-26 | 法奥意威(苏州)机器人系统有限公司 | Point cloud registration method and device based on plane fitting and electronic equipment |
| CN117745778A (en) * | 2024-02-01 | 2024-03-22 | 法奥意威(苏州)机器人系统有限公司 | Point cloud registration realization method and device, storage medium and electronic equipment |
| CN117765047B (en) * | 2024-02-22 | 2024-04-30 | 法奥意威(苏州)机器人系统有限公司 | Workpiece registration method, device, computer equipment and storage medium |
| CN117876446A (en) * | 2024-03-11 | 2024-04-12 | 法奥意威(苏州)机器人系统有限公司 | Point cloud registration method and device based on cylinder detection |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102446354A (en) * | 2011-08-29 | 2012-05-09 | 北京建筑工程学院 | Integral registration method of high-precision multisource ground laser point clouds |
| CN103295239A (en) * | 2013-06-07 | 2013-09-11 | 北京建筑工程学院 | Laser-point cloud data automatic registration method based on plane base images |
| CN104200507A (en) * | 2014-08-12 | 2014-12-10 | 南京理工大学 | Estimating method for normal vectors of points of three-dimensional point clouds |
| CN105740798A (en) * | 2016-01-27 | 2016-07-06 | 西安理工大学 | Structure analysis based identification method for object in point cloud scene |
| CN106296650A (en) * | 2016-07-22 | 2017-01-04 | 武汉海达数云技术有限公司 | A kind of laser point cloud method for registering and device |
| CN107192350A (en) * | 2017-05-19 | 2017-09-22 | 中国人民解放军信息工程大学 | A kind of three-dimensional laser scanner intrinsic parameter scaling method and device |
| CN109323656A (en) * | 2018-11-24 | 2019-02-12 | 上海勘察设计研究院(集团)有限公司 | A kind of novel target and its extraction algorithm for point cloud registering |
| CN109377521A (en) * | 2018-09-11 | 2019-02-22 | 武汉大学 | Terrestrial Laser scanner data acquire midpoint to the point cloud registration method of best fit plane |
| CN109658431A (en) * | 2018-12-26 | 2019-04-19 | 中国科学院大学 | Rock mass point cloud plane extracting method based on region growing |
| CN109685080A (en) * | 2018-12-27 | 2019-04-26 | 中国科学院大学 | Multiple dimensioned plane extracting method based on Hough transformation and region growing |
| CN109697729A (en) * | 2018-12-26 | 2019-04-30 | 中国科学院大学 | Based on the matched 3D rock mass point cloud registration method of plane polygon |
| CN109767464A (en) * | 2019-01-11 | 2019-05-17 | 西南交通大学 | A kind of point cloud registration method of low Duplication |
| CN109816664A (en) * | 2018-12-25 | 2019-05-28 | 西安中科天塔科技股份有限公司 | A kind of three-dimensional point cloud dividing method and device |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11080286B2 (en) * | 2013-12-02 | 2021-08-03 | Autodesk, Inc. | Method and system for merging multiple point cloud scans |
| CN105654422A (en) * | 2015-12-23 | 2016-06-08 | 北京观著信息技术有限公司 | Point cloud registration method and system |
| US10176589B2 (en) * | 2017-01-31 | 2019-01-08 | Mitsubishi Electric Research Labroatories, Inc. | Method and system for completing point clouds using planar segments |
| CN109272537B (en) * | 2018-08-16 | 2021-08-13 | 清华大学 | Panoramic point cloud registration method based on structured light |
-
2019
- 2019-06-24 CN CN201910550064.1A patent/CN110443836B/en active Active
Patent Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102446354A (en) * | 2011-08-29 | 2012-05-09 | 北京建筑工程学院 | Integral registration method of high-precision multisource ground laser point clouds |
| CN103295239A (en) * | 2013-06-07 | 2013-09-11 | 北京建筑工程学院 | Laser-point cloud data automatic registration method based on plane base images |
| CN104200507A (en) * | 2014-08-12 | 2014-12-10 | 南京理工大学 | Estimating method for normal vectors of points of three-dimensional point clouds |
| CN105740798A (en) * | 2016-01-27 | 2016-07-06 | 西安理工大学 | Structure analysis based identification method for object in point cloud scene |
| CN106296650A (en) * | 2016-07-22 | 2017-01-04 | 武汉海达数云技术有限公司 | A kind of laser point cloud method for registering and device |
| CN107192350A (en) * | 2017-05-19 | 2017-09-22 | 中国人民解放军信息工程大学 | A kind of three-dimensional laser scanner intrinsic parameter scaling method and device |
| CN109377521A (en) * | 2018-09-11 | 2019-02-22 | 武汉大学 | Terrestrial Laser scanner data acquire midpoint to the point cloud registration method of best fit plane |
| CN109323656A (en) * | 2018-11-24 | 2019-02-12 | 上海勘察设计研究院(集团)有限公司 | A kind of novel target and its extraction algorithm for point cloud registering |
| CN109816664A (en) * | 2018-12-25 | 2019-05-28 | 西安中科天塔科技股份有限公司 | A kind of three-dimensional point cloud dividing method and device |
| CN109658431A (en) * | 2018-12-26 | 2019-04-19 | 中国科学院大学 | Rock mass point cloud plane extracting method based on region growing |
| CN109697729A (en) * | 2018-12-26 | 2019-04-30 | 中国科学院大学 | Based on the matched 3D rock mass point cloud registration method of plane polygon |
| CN109685080A (en) * | 2018-12-27 | 2019-04-26 | 中国科学院大学 | Multiple dimensioned plane extracting method based on Hough transformation and region growing |
| CN109767464A (en) * | 2019-01-11 | 2019-05-17 | 西南交通大学 | A kind of point cloud registration method of low Duplication |
Non-Patent Citations (3)
| Title |
|---|
| Automated coarse registration of point clouds in 3d urban scenes using voxel based plane constraint;Y. Xu 等;《ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences》;20170922;第185-191页 * |
| 激光扫描匹配方法研究综述;宗文鹏 等;《中国光学》;20181212;第11卷(第06期);第914-930页 * |
| 采用八叉树体素生长的点云平面提取;李明磊 等;《光学精密工程》;20180115;第26卷(第01期);第172-183页 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11787051B2 (en) * | 2021-11-25 | 2023-10-17 | Nanjing University Of Aeronautics And Astronautics | Method for automatically processing structure-reinforcing member of aircraft |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110443836A (en) | 2019-11-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110443836B (en) | Point cloud data automatic registration method and device based on plane features | |
| Yang et al. | Automated registration of dense terrestrial laser-scanning point clouds using curves | |
| CN104200461B (en) | The remote sensing image registration method of block and sift features is selected based on mutual information image | |
| Kang et al. | Automatic targetless camera–lidar calibration by aligning edge with gaussian mixture model | |
| CN103727930A (en) | Edge-matching-based relative pose calibration method of laser range finder and camera | |
| CN111612850B (en) | Point cloud-based pig body ruler parameter measurement method | |
| US20150199573A1 (en) | Global Scene Descriptors for Matching Manhattan Scenes using Edge Maps Associated with Vanishing Points | |
| Urban et al. | Finding a good feature detector-descriptor combination for the 2D keypoint-based registration of TLS point clouds | |
| Chen et al. | Robust affine-invariant line matching for high resolution remote sensing images | |
| CN110796691B (en) | Heterogeneous image registration method based on shape context and HOG characteristics | |
| CN110569861A (en) | Image matching positioning method based on point feature and contour feature fusion | |
| CN107220996A (en) | A kind of unmanned plane linear array consistent based on three-legged structure and face battle array image matching method | |
| CN107610174B (en) | Robust depth information-based plane detection method and system | |
| Wang | Automatic extraction of building outline from high resolution aerial imagery | |
| Lv et al. | Automatic registration of airborne LiDAR point cloud data and optical imagery depth map based on line and points features | |
| CN105654479A (en) | Multispectral image registering method and multispectral image registering device | |
| Wang et al. | Point based registration of terrestrial laser data using intensity and geometry features | |
| Parmehr et al. | Automatic registration of optical imagery with 3d lidar data using local combined mutual information | |
| Parmehr et al. | Automatic parameter selection for intensity-based registration of imagery to LiDAR data | |
| Sui et al. | Automatic image registration based on shape features and multi-scale image segmentation | |
| CN112767457A (en) | Principal component analysis-based plane point cloud matching method and device | |
| CN110969650B (en) | Intensity image and texture sequence registration method based on central projection | |
| CN107765257A (en) | A kind of laser acquisition and measuring method based on the calibration of reflected intensity accessory external | |
| Zhong et al. | Triple screening point cloud registration method based on image and geometric features | |
| Shen et al. | A 3D modeling method of indoor objects using Kinect sensor |
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 |