CN114463396B - Point cloud registration method utilizing plane shape and topological graph voting - Google Patents
Point cloud registration method utilizing plane shape and topological graph voting Download PDFInfo
- Publication number
- CN114463396B CN114463396B CN202210012942.6A CN202210012942A CN114463396B CN 114463396 B CN114463396 B CN 114463396B CN 202210012942 A CN202210012942 A CN 202210012942A CN 114463396 B CN114463396 B CN 114463396B
- Authority
- CN
- China
- Prior art keywords
- plane
- point
- contour
- planes
- 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
Links
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 30
- 230000009466 transformation Effects 0.000 claims abstract description 13
- 239000013598 vector Substances 0.000 claims abstract description 7
- 238000000357 thermal conductivity detection Methods 0.000 claims description 34
- 238000005070 sampling Methods 0.000 claims description 21
- 230000011218 segmentation Effects 0.000 claims description 19
- 238000004364 calculation method Methods 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 3
- 238000007670 refining Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000011426 transformation method Methods 0.000 claims description 3
- 241000764238 Isis Species 0.000 claims description 2
- 238000012163 sequencing technique Methods 0.000 claims 1
- 238000013461 design Methods 0.000 abstract description 5
- 230000008569 process Effects 0.000 abstract description 3
- 238000005516 engineering process Methods 0.000 description 5
- 238000010276 construction Methods 0.000 description 3
- 238000013507 mapping Methods 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 238000013519 translation Methods 0.000 description 2
- 238000007792 addition Methods 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000013178 mathematical model Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
Classifications
-
- 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
- G06T19/00—Manipulating 3D models or images for computer graphics
- G06T19/20—Editing of 3D images, e.g. changing shapes or colours, aligning objects or positioning parts
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Architecture (AREA)
- Computer Graphics (AREA)
- Computer Hardware Design (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a point cloud registration method utilizing plane shapes and topology graph voting, which comprises the following steps: step 1, roughly dividing point cloud into plane elements by adopting a voxelized point cloud plane dividing algorithm, and calculating accurate plane parameters by using a RANSAC plane fitting method; step 2, selecting a plane shape as a plane matching feature, and designing a plane shape descriptor to perform homonymous plane matching; step 3, constructing a topological graph according to the matched relationship between the positions of the same-name planes and the angles of normal vectors, and eliminating the mismatching planes based on a same-name plane edge voting strategy; and 4, selecting optimal groups of homonymous planes to calculate transformation matrixes of the two-site cloud to finish the registration of the point cloud. The invention designs the HIA-TCD plane shape descriptor, which considers the holes on the plane and the phenomenon of partial shielding in the scanning process, and can improve the robustness of plane matching.
Description
Technical Field
The invention relates to a point cloud automatic coarse registration algorithm based on plane shape matching and topological graph voting, and belongs to the field of remote sensing mapping.
Background
In recent years, with the development of 3D Lidar technology, point clouds have been widely used in urban scenes, such as three-dimensional reconstruction of buildings, urban planning, indoor navigation, and the like, and the point clouds have proven to be one of the most suitable data sources for drawing urban scenes. The method is limited by the scanning distance and the scanning mode of the three-dimensional laser scanner, the complete scene data can not be obtained through one-station scanning, the data integrity and the quality are improved through multiple-angle, multiple-platform, multiple-measuring-station and multiple-time equal methods, and the point clouds of different measuring stations are unified to the same coordinate system through a registration technology, so that the point cloud registration has an important position in the laser radar research and is an important basis of other point cloud processing technologies. The research is widely distributed in the fields of computer vision, photogrammetry, forest investigation, robot technology and the like.
The point cloud registration is always a research hotspot in the three-dimensional laser field, and the process of the point cloud registration is to calculate rigid transformation parameters between two site clouds, take one site cloud as a target point cloud, and convert the other site (source point cloud) into the same coordinate system as the target point cloud. A rigid transformation is a low-dimensional entity that can be uniquely recovered as long as the correct correspondence is known between the three point pairs. The fine registration algorithm represented by the ICP algorithm is one of effective methods for solving the registration problem, and the algorithm repeatedly selects corresponding relation point pairs and calculates an optimal rigid body transformation until the convergence accuracy requirement of correct registration is satisfied. But this algorithm requires an initial value for the given rotation matrix, otherwise it is easy to get into local optima. Thus, coarse registration of the two-site cloud is required to obtain the initial value. The point cloud coarse registration technology is further divided into a point-based method, a line-based method and a surface-based method according to the primitive difference of feature extraction. Wherein the point-based method is less efficient and less accurate and is susceptible to noise points. Although the line-based method can improve the efficiency, the homonym is difficult to match, and the use scene is limited. The plane-based point cloud registration algorithm can greatly reduce the number of matching primitives, can accelerate the matching efficiency of the primitives with the same name, on the other hand, the characteristics of the planes are more stable, the Xiao et al calculate the area of the planes after dividing the planes by using the area growth algorithm, then find two groups of planes with the approximately equal area to calculate the rotation quantity, if the rotation is true, find a third group of planes, judge the consistency according to the normal, and finally calculate the transformation parameters. Dold et al match based on the area of the plane, the length of the boundary, the border frame, and the average length as descriptors. Li et al find the corresponding match in the two-site cloud through a special angle between the three planes. However, due to the influence of the scanning position, the building plane point cloud obtained by the laser scanner is usually shielded, and the matching of the same-name planes by means of the feature constraints such as the area of the planes, bounding boxes, vector angles and the like is not reliable.
Disclosure of Invention
The main technical problems to be solved by the invention are as follows: an accurate and robust descriptor is designed aiming at plane characteristics and is used for matching with a same-name plane so as to solve the problem of registering a large data volume point cloud in a city scene.
Aiming at the problems, the invention provides a method for solving rough registration of urban market scenic spot clouds based on plane shape and topological graph voting. In human perception rules, whether two planes in a city scene are homonymous planes is generally judged according to the similarity of the plane shapes and the positional relationship between the planes, the plane shapes are used as geometrical matching features according to the principle, the homonymous planes are matched through designing plane shape descriptors, then in the matched homonymous planes, a topological graph is constructed according to the positions and normal features of the planes, and the false matching is removed according to the voting principle. And finally, calculating rigid transformation parameters between the two site clouds by selecting three pairs of optimal matching planes.
In order to achieve the above purpose, the technical scheme provided by the invention is as follows:
step 1, roughly dividing point cloud into plane elements by adopting a voxelized point cloud plane dividing algorithm, and calculating accurate plane parameters by using a RANSAC plane fitting method;
step 2, selecting a plane shape as a plane matching feature, and designing a plane shape descriptor to perform homonymous plane matching;
step 3, constructing a topological graph according to the matched relationship between the positions of the homonymous planes and the normal vector included angles, and eliminating the mismatching planes based on the homonymous plane edge voting strategy;
and 4, selecting optimal groups of homonymous planes to calculate transformation matrixes of the two-site cloud to finish the registration of the point cloud.
Further, in step 1, firstly, a binary-coded octree is constructed to complete the voxelization of the point cloud, then, the characteristics of each voxel are calculated, and on the basis, a region growing strategy is adopted to segment a plane: selecting voxels with obvious plane characteristics as seeds, merging adjacent voxels with similar plane characteristics with the seeds into a cluster, and refining the undivided voxels to improve the segmentation result; finally, acquiring accurate plane parameters by using a RANSAC plane fitting method;
further, the specific implementation of the step 2 comprises the following sub-steps;
step 2.1, calculating the inner angle HIA shape context considering the vertical plane holes;
step 2.2, calculating the triangle centroid distance TCD shape context;
step 2.3, calculating an HIA-TCD shape descriptor by combining the internal angle and the triangle centroid distance;
step 2.4, calculating the similarity of two plane contour shapes based on the HIA-TCD shape descriptor;
and 2.5, carrying out homonymous plane matching based on the plane shape similarity.
Further, the specific implementation manner of step 2.1 is as follows;
for a given planar shapeForm O, first rotate the segmented plane to xo y The method comprises the steps of extracting a contour by a plane, wherein the contour comprises an outer contour and a hole contour, uniformly sampling the number of points of the contour, and obtaining an outer contour sampling point set P= { P 1 ,p 2 ,...,p n Arbitrary two points p i ,p j The internal angle HIA between the two points taking into account the holes is defined as: contour point p i Is defined by the sum p of the contour tangents of (2) i To p j Shortest path ψ (p i ,p j Angle Ang (p) between the O-directions i ,p j ) Shortest path ψ (p i ,p j The limiting condition of O) is that (1) the interior (2) of the plane outline cannot pass through the holes on the plane, and the route from each point to other points can be obtained through an SPFA shortest path query algorithm; the outline extraction method comprises the following steps: and setting edge constraint to construct a triangular net by using a divide-and-conquer Delaunay algorithm, and respectively finding out an outer contour and a hole contour according to the characteristic that only one triangle is connected to the contour edge.
Further, the specific implementation manner of step 2.2 is as follows;
given plane outline sampling point set p= { P 1 ,p 2 ,...,p n P is }, where i ＝(x i ,y i ,z i ) Representing points and three-dimensional coordinates thereof, wherein n represents the number of sampling points, and the sampling points are ordered anticlockwise; first, a point set center g= (x) G ,y G ,z G ) For each contour point p i ＝(x i ,y i ,z i ) (i=1, 2,.,. N), N being the number of points, another point p on the set of contour points can be found j ＝(x j ,y j ,z j ) (j=1, 2,..n, i+.j), these two points and the contour center point make up a triangle Δp i Gp j Then using any two points p on the contour i 、p j And the contour center point G form a triangle delta p i Gp j Then calculate the center point of this triangleThe calculation is as follows:
for each contour point p i N-1 triangles can be obtained, and then the point p is calculated i And a center point g ij Distance of (2)This feature is referred to as a three-dimensional triangle centroid distance.
Further, the specific implementation manner of step 2.3 is as follows;
given plane outline sampling point set p= { P 1 ,p 2 ,...,p n N is the number of points on the contour, for a point p on the contour i Firstly, calculating the internal angle HIA and the triangle centroid distance TCD of the hole formed by the point and other points:
the characteristics of HIAs and TCDs for all contour points of a certain plane O can be obtained as follows:
wherein TCDs (O) is a matrix of (N-1) x N, and the ith column is the sampling point p i TCDs features with other points; since the shape descriptions obtained from distance are not scale-invariant, TCDs (O) are normalized by dividing by the maximum absolute value of each row, as follows:
wherein Dist' (p) i ,g ij ) Representing normalized post-point p i And a center point g ij Is a distance of (2); for point p i In constructing HIA-TCD shape context descriptor histogram h i When using n a The number of statistical squares representing HIA is taken as the x-axis, n t The number of square lattices representing TCD as y-axis; combining the two into a 2D histogram according to p i HIA (p) of dot i ) And normalized TCD (p i ) And calculating the index of each value falling in the histogram interval, counting the number m of each interval, and obtaining the value of the two-dimensional histogram bin according to m/(N-1).
Further, the specific implementation manner of step 2.4 is as follows;
let p be i Is the point on the first planar contour, q j Is a point on the second planar contour, then the shape context similarity C of the two points ij ＝C(p i ,q j ) According to χ 2 The test statistic is calculated as follows:
wherein h is P,i (k) A contour point P being a plane P i HIA-TCD shape context descriptor histogram, h Q,j (k) The contour point Q being the plane Q j The HIA-TCD shape context descriptor histogram, K representing the number of histogram bins; the shape matching cost formula is as follows:
where pi (i) represents the number of points where the match was successful, the match pi for the planar profiles P and Q is a mapping from 1,2, n to 1,2, m, if pi (i) noteq0 represents P i And q π(i) Is a match, the optimal match is defined as pi so that the matching cost H (pi) is minimum, and the number of points successfully matched is obtained by using a dynamic programming matching algorithm.
Further, the specific implementation manner of step 2.5 is as follows;
two-site clouds S and T are known, S is a source point cloud, T is a target point cloud, and the segmented plane set is s= { S 1 ,s 2 ,...,s m Sum t= { T 1 ,t 2 ,...,t n T= { T } 1 ,t 2 ,...,t n Plane t in } i And S= { S 1 ,s 2 ,...,s m Plane s in } j The matching cost of the two can be calculated according to the matching cost formula ij And the number N of points successfully matched matched H (pi) and pi (i) in the step 2.4 are respectively corresponding; from the matching result, the higher the similarity of the two plane contours, the matching cost of the two plane contours ij The smaller the size; matching the point cost pcost matched ＝cost ij /N matched Selecting an optimal corresponding plane as a comparison value, and selecting S= { S 1 ,s 2 ,...,s m Intermediate and t i Plane t with minimum matching point cost j Obtain the plane pair (t) i ,s j ) Finally, T= { T is calculated 1 ,t 2 ,...,t n The homonymous planes of all planes in a } form a homonymous plane set m= { (t) i ,s i )|t i ∈T；s i ∈S；i＝1,2,...n}。
Further, the specific implementation manner of the step 3 is as follows;
first, according to the homonymous plane set m= { (t) i ,s i )|t i ∈T；s i E S; i=1, 2,..n } build topology G S (V S ,E S ) And G T (V T ,E T ) The method comprises the steps of carrying out a first treatment on the surface of the Wherein V is S (V T ) Is a node of the topology, each node representing a plane s in M i (t i ) Its attribute is defined by plane s i (t i ) Center point coordinates of (2)Representing that the properties of an edge include two parts: distance of two nodes:and the normal angle between the planes represented by the two nodesWherein->Refers to the coordinates of two nodes, n i ，n j Refers to the normal of the plane;
since M is a set of co-name planes, therefore,for the node correspondence set, if->Andare correctly matched, then their edge pairs +.>Is also a correct match and in the ideal case +.>Wherein V represents a node set, V represents a node in the set, E represents a set of edges, E represents an element in the set of edges, E (i,j) Representing the edge formed by connecting the nodes i and j; wherein (1)>Represents->Length of->Represents->I.e. the angle between the two planes, if +.>Andthe relation of (2) is represented by->And->At least one of which is a false match; in practice, however, only +.>And->The difference in length and angle is small, and the two are considered to be correct matches.
If it isIs the correct corresponding edge set, consider +.>And->The two corresponding planes with the same name are correctly matched, and then the node set corresponding to the two planes is +.>And->Each 1-vote, this principle is called a plane relation side voting strategy; obviously, a certain node pair ∈ ->Is +.>The more the group of homonymous plane correspondences are, the more robust the group of homonymous plane correspondences are; if the ticket number of a certain node is lower than N ε The node is considered to be the wrong homonymous plane match and is taken from M = { (t) i ,s i )|t i ∈T；s i E S; i=1, 2,..n } culling, and finally sorting the nodes in M according to the vote count of each node.
Further, the specific implementation manner of the step 4 is as follows:
selecting three groups of optimal matching planes for calculation, selecting the ground as a first group of homonymous planes according to the constraint condition of maximum area or maximum points or minimum Z coordinates of central points in Ping Miandian cloud, and then collecting M = { (t) of homonymous planes subjected to voting strategies i ,s i )|t i ∈T；s i E S; and selecting two groups of non-parallel planes to form three groups of homonymous planes according to the number of votes by the nodes in i=1, 2, & gt. N } and finally finishing point cloud registration by using a closed form solution estimation rigid body transformation method according to plane parameters of the three groups of homonymous planes.
Compared with the prior art, the invention has the advantages and beneficial effects that:
1) A new method of coarse registration of planar primitives is designed. The plane shape descriptor is used for matching the same name plane, compared with a descriptor (descriptor) of the point feature, the descriptor is less affected by noise, and the feature is more stable; the planar properties such as area, bounding box, etc. are selected as descriptions in other planar feature-based registration algorithms to be more robust.
2) In order to improve the matching capability of incomplete planes, the invention designs the HIA-TCD plane shape descriptor, which considers the holes on the planes and the phenomenon of partial shielding in the scanning process, and can improve the robustness of plane matching.
3) In order to select the optimal matching plane, the optimal matching problem is converted into an edge matching problem based on the topological graph representation. The invention establishes a mathematical model with maximized edge correspondence. On the basis of matched planes with the same name, a topological graph is constructed according to the position and normal vector relation of the planes, an edge voting strategy is used, incorrect matching is removed, and three groups of optimal matching calculation rigid body transformation matrixes are selected.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a schematic view of calculating an inner distance and an inner angle of an elevation hole according to an embodiment of the present invention.
Fig. 3 is a schematic diagram of triangle centroid distance calculation in an embodiment of the invention.
Fig. 4 is a schematic diagram of a matching result between two planar shape contour points in an embodiment of the present invention.
Detailed Description
The technical scheme of the invention is further described below with reference to the accompanying drawings and examples.
Three steps in the plane-based registration method are the most critical:
(1) How to extract planes quickly and accurately, in the current plane-based registration, planes in a scene are extracted mostly by using RANSAC or variants of RANSAC, but this method has low efficiency of plane segmentation, and is not applicable to the point cloud plane segmentation task of a large urban scene. Other planar segmentation methods such as region growing and Hough transformation can be used for planar segmentation, but the efficiency of the planar segmentation method is difficult to meet the requirement of planar segmentation of a large scene, and the accuracy of planar parameter calculation is affected by the existence of over-segmentation phenomenon. The voxel-based plane segmentation algorithm can efficiently complete the task of large-scene point cloud plane segmentation, but saw-tooth under-segmentation results can be generated at the plane edge.
(2) And (5) extracting the plane characteristics and matching the corresponding relation. The key to the plane-based registration method is the matching of the correct homonymous planes. In the existing registration method based on plane features, plane attributes such as areas, bounding boxes or special angles between plane normals are generally selected as descriptors to find corresponding matches in the two-site cloud. The descriptors are not robust by the effects of occlusion.
(3) And eliminating mismatching homonymous planes. There are many similar, symmetrical or repetitive structures in urban scenes that to some extent are prone to mismatching between planes. The existing method generally utilizes an exhaustive search method to establish the corresponding relation between different scanning planes, and eliminates incorrect homonymous planes, so that the algorithm complexity of the method is too high. Or the method for eliminating the mismatching based on the points is similar to the method for eliminating the mismatching based on the points, and the method is not stable enough.
The invention aims to design a point cloud registration algorithm suitable for urban scenes, and firstly, the algorithm can carry out rapid plane segmentation and accurate parameter estimation on the urban scene point cloud with large data quantity. Secondly, a robust plane descriptor is designed, the descriptor has good robustness, and the descriptor can be accurately matched with a same-name plane in two-station scanning even if the point cloud is shielded. And finally, eliminating mismatching homonymous planes according to the geometrical relationship of the matched planes and the human perception rule, and selecting the optimal three groups of homonymous planes to finish point cloud registration.
The invention is generally divided into three steps: in the first step, in order to improve the efficiency of plane segmentation and the accuracy of parameter estimation, a binary-coded octree efficient voxelized point cloud plane segmentation algorithm is used to coarsely segment the point cloud into plane primitives and a RANSAC plane fitting method is used to calculate accurate plane parameters. And secondly, selecting a plane shape as a plane matching characteristic, and providing a space plane shape descriptor combining an internal angle with a triangle centroid distance (HIA-TCD), wherein the descriptor can still be correctly matched with a same-name plane even if the plane contour is incomplete. And finally, constructing a topological graph according to the matched relationship between the positions of the homonymous planes and the normal vector included angles, designing a homonymous plane edge voting strategy, eliminating the mismatching planes, and selecting the optimal three groups of homonymous planes to calculate a transformation matrix of the two-site cloud to finish the rough registration of the point cloud. The whole technical flow is shown in the following chart
Step 1, roughly dividing point cloud into plane elements by adopting a voxelized point cloud plane dividing algorithm, and calculating accurate plane parameters by using a RANSAC plane fitting method;
in order to improve the efficiency of large-scale point cloud segmentation, the invention uses an efficient plane segmentation algorithm based on encoding voxels to coarsely segment the point cloud into planes. Firstly, constructing a binary coding octree to complete voxelization of the point cloud, and then calculating the characteristics of each voxel. On the basis, a region growing strategy is adopted to divide the plane: voxels with significant planar features are selected as seeds and then neighboring voxels with similar planar features to the seeds are merged into one cluster. And finally, refining the undivided voxels to improve the segmentation result. In voxel-based planar segmentation, the arrangement of larger voxel nodes can improve the efficiency of planar segmentation, but can lose the accuracy of planar parameters. The invention uses the RANSAC plane fitting method to improve the parameter precision. And obtaining the inner points of the fitting model by setting smaller fitting distance, and then iteratively estimating the optimal plane parameters. Because the plane is re-fitted on the basis of the segmented rough plane, the RANSAC iteration times are less, and the overall segmentation efficiency is still higher.
Step 2, selecting a plane shape as a plane matching feature, and performing homonymous plane matching based on a plane shape descriptor
The plane in the structural building can be regarded as a 3d shape with 2d characteristics, and the key of shape matching is to construct a shape descriptor which has rich characteristic information and rotation, translation, scale invariance and the like and meets the matching requirement. In addition to the foregoing requirements, holes in the building plane are also considered as well as partial shielding. The invention designs the HIA-TCD shape descriptor for fusing the angles in the contour and the distances between the centers of mass of the triangles according to the requirements.
Step 2.1, calculating an interior angle (HIA) shape context taking into account the elevation holes
First, a definition of the inner distance is given: outer contour sampling point set p= { P for given planar shape O 1 ,p 2 ,...,p n Two points p are taken i ,p j The internal distance between two points is defined as the internal connection node of shape OShortest path ψ (p i ,p j Length of O) by d (p) i ,p j O). According to the definition of the inner distance, the inner angle is defined as: contour point p i Is defined by the contour tangent and the corresponding point p j Shortest path ψ (p i ,p j Angle between the O) directions. As shown in fig. 2 (f). The specific calculation method considering the internal distance (HID) and the internal angle (HIA) of the vertical hole is shown in fig. 2, and the specific steps are as follows:
(1) Firstly, the segmented plane is rotated to an xoy plane, then, a dividing and controlling Delaunay algorithm is used for setting edge constraint to construct a triangular net, and the external contour and the hole contour are respectively found out in sequence according to the characteristic that only one triangle is connected to the contour edge, wherein the result is shown in fig. 2 (b).
(2) After the contour is found, the number of points of the contour is uniformly sampled, the number of sampling points of the outer contour is set to be 100, and the number of sampling points of the contour of the hole is 50 as shown in fig. 2 (c).
(3) According to the sampling result of the outer contour, constructing a graph structure taking the inner holes into consideration, and considering two aspects in construction: first is the contour point p i And p is as follows j The connection line between two points is in the interior of the plane outline, and the connection line between two points cannot pass through the holes on the vertical surface as shown in fig. 2 (d). In order to improve the shortest path query efficiency, an SPFA shortest path query algorithm is used to obtain the route from each point to other points, and the internal angle Ang (p i ,p j )。
Step 2.2, calculating Triangle Centroid Distance (TCD) shape context
The invention designs a space Triangle Centroid Distance (TCD) shape context to improve the matching performance of a partially-shielded shape, and the construction method is as follows: as shown in fig. 3, a given out-of-plane contour sample point set p= { P 1 ,p 2 ,...,p n P is }, where i ＝(x i ,y i ,z i ) Representing the points and their three-dimensional coordinates, n representing the number of sampling points, the sampling points being ordered in a counter-clockwise manner. First, a point set center g= (x) G ,y G ,z G ) For each contour point p i ＝(x i ,y i ,z i ) (i=1, 2,., N), another point p on the set of contour points can be found j ＝(x j ,y j ,z j ) (j=1, 2,..n, i+.j), these two points and the contour center point make up a triangle Δp i Gp j Then calculate the center point of this triangleThe calculation is as follows:
for each contour point p i N-1 triangles can be obtained, and then the point p is calculated i And a center point g ij Distance of (2)This feature is referred to as a three-dimensional triangle centroid distance.
Step 2.3, combining the internal Angle (HIA) and Triangle Centroid Distance (TCD), calculating the HIA-TCD shape descriptor
The method is more beneficial to the description of the shape by combining the distance and the angle of the center of mass of the triangle with the inner angle of the hole, and the HIA-TCD shape context descriptor is designed, so that the HIA-TCD shape context descriptor has the matching capability of global features and local features of a space plane, the holes, the shielding and the like. The construction method comprises the following steps:
given plane outline sampling point set p= { P 1 ,p 2 ,...,p n N is the number of points on the contour, for a point p on the contour i Firstly, calculating the internal angle HIA and the triangle centroid distance TCD of the hole formed by the point and other points:
the characteristics of HIAs and TCDs for all contour points of a certain plane O can be obtained as follows:
wherein TCDs (O) is a matrix of (N-1) x N, and the ith column is the sampling point p i TCDs features with other points. Since the shape descriptions obtained from distance are not scale-invariant, TCDs (O) are normalized by dividing by the maximum absolute value of each row, as follows:
wherein Dist' (p) i ,g ij ) Representing normalized post-point p i And a center point g ij Is a distance of (2); for point p i In constructing HIA-TCD shape context descriptor histogram h i When using n a The number of statistical square lattices (bins) representing HIA (1D) (taking a fixed value of n a =12) as x-axis, n t Representing the number of bins of TCD (1D) (taking a fixed value n t =5) as y-axis. Combining the two into a 2D histogram according to p i HIA (p) of dot i ) And normalized TCD (p i ) And calculating the index of each value falling in the histogram interval, counting the number m of each interval, and obtaining the value of the two-dimensional histogram bin according to m/(N-1).
Step 2.4, calculating the similarity of the profiles of the two plane shapes based on the HIA-TCD shape descriptor
Let p be i Is the point on the first planar contour, q j Is a point on the second planar contour, then the shape context similarity C of the two points ij ＝C(p i ,q j ) According to χ 2 The test statistic is calculated as follows:
wherein h is P,i (k) A contour point P being a plane P i Shape context descriptor histogram of h Q,j (k) The contour point Q being the plane Q j And K represents the number of histogram bins. In a general feature matching algorithm, an optimal matching algorithm is generally applied to obtain optimal matching, a minimum cost is obtained through solving, and finally all shape matching costs can be obtained based on the optimal matching obtained through solving, wherein a shape matching cost formula is as follows:
the match pi (i) of planes P and Q is a mapping from 1,2, n to 1,2, m, if pi (i) noteq0 represents P i And q π(i) Is matched. The optimal match is defined as pi (i) such that the matching cost H (pi) is minimized. For the project specific problem, when the shape contour sequence information is known, the ordered contour matching is more accurate and efficient by using a Dynamic Programming (DP) algorithm, so that in order to solve the matching problem, a shape matching cost function is solved by using the dynamic programming matching algorithm, and the successfully matched point pi (i) is obtained. The result of the matching of the two planar shapes is shown in fig. 4.
Step 2.5, matching the same name plane based on the similarity of the plane shapes
Two-site clouds (source point cloud) and T (target point cloud) are known, and the segmented plane set thereof is s= { S 1 ,s 2 ,...,s m Sum t= { T 1 ,t 2 ,...,t n T= { T } 1 ,t 2 ,...,t n Plane t in } i And S= { S 1 ,s 2 ,...,s m Plane s in } j The matching cost of the two can be calculated according to the matching cost formula ij And the number N of points successfully matched matched From the matching results, it can be seen that the higher the similarity of the two plane profiles, the higher their matching cost, corresponding to H (pi) and pi (i) in step 2.4, respectively ij The smaller. The invention will matchPoint cost pcost matched ＝cost ij /N matched And selecting an optimal corresponding plane as a comparison value. Select s= { S 1 ,s 2 ,...,s m Intermediate and t i Plane t with minimum matching point cost j Obtain the plane pair (t) i ,s j ) Finally, T= { T is calculated 1 ,t 2 ,...,t n The homonymous planes of all planes in a } form a homonymous plane set m= { (t) i ,s i )|t i ∈T；s i ∈S；i＝1,2,...n}。
Step 3, constructing a topological graph according to the matched homonymous plane position and normal vector included angle relation, and based on a homonymous plane edge voting strategy, rejecting the mismatching planes, which are very similar in shape and usually have repeated structures and parallel relations due to the fact that a great number of plane elements are contained in the urban scene point cloud, so that the mismatching of the planes is very easy to cause, and therefore, the mismatching homonymous planes need to be rejected, and the most stable homonymous planes are selected for registration. First, according to the homonymous plane set m= { (t) i ,s i )|t i ∈T；s i E S; i=1, 2,..n } build topology G S (V S ,E S ) And G T (V T ,E T ) V represents node set, V represents node in the set, E represents graph edge set, E represents one element in the edge set, one edge connects two nodes, so E is used (i,j) And (3) representing. Wherein V is S (V T ) Is a node of the topology, each node representing a plane s in M i (t i ) Its attribute is defined by plane s i (t i ) Center point coordinates of (2)To reduce the effects of noise and point cloud loss, only the x, y coordinates are used; e (E) S (E T ) For an edge of a topology graph, the properties of the edge include two parts: distance between two nodes->AndThe angle between two nodesWherein->Refers to the coordinates of two nodes, n i ，n j Refers to the normal to the plane.
Since M is a set of co-name planes, therefore,for the node correspondence set, if->Andare correctly matched, then their edge pairs +.>Is also a correct match and in the ideal case +.>Wherein (1)>Represents->Length of->Represents->I.e. the angle between two planes. If it does not meet +.>And->Is represented in the relation ofAnd->At least one of which is a false match.
The principle is based on the fact that two identical name planes are in perfect equalityAnd->The difference in length and angle is small, and the two are considered to be correct matches. If->Is the correct corresponding edge set, consider +.>And->The two corresponding planes with the same name are correctly matched, and then the node set corresponding to the two planes is +.>And->Each vote 1, this principle is referred to as a planar relationship side voting strategy. Obviously, a certain node pair ∈ ->Is +.>The more the same name plane correspondence is, the more robust the set. If the ticket number of a certain node is lower than N ε The node is considered to be the wrong homonymous plane match and is taken from M = { (t) i ,s i )|t i ∈T；s i E S; i=1, 2,..n } culling, and finally sorting the nodes in M according to the vote count of each node.
Step 4, selecting three optimal groups of homonymous planes to calculate transformation matrixes of two-site clouds to finish coarse registration of the point clouds
The conversion between two site clouds S and T typically consists of a 3D rotation parameter R and a 3D translation parameter T, which for plane-based registration methods typically require three accurate co-named planes to directly calculate the transformation parameters R and T. In order to reduce the number of the same name planes to adapt to the situation of fewer planes in a scene, three groups of optimal matching planes are selected for calculation. In urban scenes, the ground is a plane with stable characteristics and easy recognition, so that the ground points are selected as a first group of homonymous planes, the ground is easy to select from the segmented planes, and available characteristics such as the largest area, the largest number of points, the lowest Z coordinate of a central point and the like are available. Then for the same name plane set m= { (t) that has been voting strategy i ,s i )|t i ∈T；s i E S; the nodes in i=1, 2..n } select two sets of non-parallel planes from large to small according to the number of votes, and they form three sets of identical name planes with the ground. Finally, a closed form solution estimation rigid body transformation method is used according to plane parameters of three groups of homonymous planes [1] And (5) finishing point cloud registration.
[1].Khoshelham,K.."Closed-form solutions for estimating a rigid motion from plane correspondences extracted from point clouds."Isprs Journal of Photogrammetry&Remote Sensing 114.Apr.(2016):78-91.
The specific embodiments described herein are offered by way of example only to illustrate the spirit of the invention. Those skilled in the art may make various modifications or additions to the described embodiments or substitutions thereof without departing from the spirit of the invention or exceeding the scope of the invention as defined in the accompanying claims.
Claims (4)
1. A point cloud registration method using plane shape and topology voting, comprising the steps of:
step 1, roughly dividing point cloud into plane elements by adopting a voxelized point cloud plane dividing algorithm, and calculating accurate plane parameters by using a RANSAC plane fitting method;
step 2, selecting a plane shape as a plane matching feature, and designing a plane shape descriptor to perform homonymous plane matching;
the specific implementation of the step 2 comprises the following sub-steps;
step 2.1, calculating the inner angle HIA shape context considering the vertical plane holes;
the specific implementation mode of the step 2.1 is as follows;
for a given plane shape O, firstly, rotating a segmented plane to an xoy plane, extracting a contour, including an outer contour and a hole contour, uniformly sampling the number of points of the contour, and obtaining an outer contour sampling point set P= { P 1 ,p 2 ,...,p N Arbitrary two points p i ,p j The internal angle HIA between the two points taking into account the holes is defined as: contour point p i Is defined by the sum p of the contour tangents of (2) i To p j Shortest path ψ (p i ,p j Angle Ang (p) between the O-directions i ,p j ) Shortest path ψ (p i ,p j The limiting condition of O) is that (1) the interior (2) of the plane outline cannot pass through the holes on the plane, and the route from each point to other points can be obtained through an SPFA shortest path query algorithm; the outline extraction method comprises the following steps: setting edge constraint to construct a triangular net by using a divide-and-conquer Delaunay algorithm, and respectively finding out an outer contour and a hole contour according to the characteristic that only one triangle is connected to the contour edge in sequence;
step 2.2, calculating the triangle centroid distance TCD shape context;
the specific implementation mode of the step 2.2 is as follows;
given plane outline sampling point set p= { P 1 ,p 2 ,...,p N P is }, where i ＝(x i ,y i ,z i ) Representing points and three-dimensional coordinates thereof, and sequencing sampling points according to anticlockwise; first, a point set center g= (x) G ,y G ,z G ) For each contour point p i ＝(x i ,y i ,z i ) (i=1, 2,.,. N), N being the number of points, another point p on the set of contour points can be found j ＝(x j ,y j ,z j ) (j=1, 2,..n, i+.j), these two points and the contour center point make up a triangle Δp i Gp j Then using any two points p on the contour i 、p j And the contour center point G form a triangle delta p i Gp j Then calculate the center point of this triangleThe calculation is as follows:
for each contour point p i N-1 triangles can be obtained, and then the point p is calculated i And a center point g ij Distance of (2)This feature is referred to as a three-dimensional triangle centroid distance;
step 2.3, calculating an HIA-TCD shape descriptor by combining the internal angle and the triangle centroid distance;
the specific implementation mode of the step 2.3 is as follows;
given plane outline sampling point set p= { P 1 ,p 2 ,...,p N N is the number of points on the contour, for a point p on the contour i Firstly, calculating the internal angle HIA and the triangle centroid distance TCD of the hole formed by the point and other points:
the characteristics of HIAs and TCDs for all contour points of a certain plane O can be obtained as follows:
wherein TCDs (O) is a matrix of (N-1) x N, and the ith column is the sampling point p i TCDs features with other points; since the shape descriptions obtained from distance are not scale-invariant, TCDs (O) are normalized by dividing by the maximum absolute value of each row, as follows:
wherein Dist' (p) i ,g ij ) Representing normalized post-point p i And a center point g ij Is a distance of (2); for point p i In constructing HIA-TCD shape context descriptor histogram h i When using n a The number of statistical squares representing HIA is taken as the x-axis, n t The number of square lattices representing TCD as y-axis; combining the two into a 2D histogram according to p i HIA (p) of dot i ) And normalized TCD (p i ) Calculating the index of each value falling in the histogram interval, counting the number m of the values falling in each interval, and obtaining the value of the two-dimensional histogram bins according to m/(N-1);
step 2.4, calculating the similarity of two plane contour shapes based on the HIA-TCD shape descriptor;
the specific implementation mode of the step 2.4 is as follows;
let p be i Is the firstPoints on a planar contour, q j Is a point on the second planar contour, then the shape context similarity C of the two points ij ＝C(p i ,q j ) According to χ 2 The test statistic is calculated as follows:
wherein h is P,i (k) A contour point P being a plane P i HIA-TCD shape context descriptor histogram, h Q,j (k) The contour point Q being the plane Q j The HIA-TCD shape context descriptor histogram, K representing the number of histogram bins; the shape matching cost formula is as follows:
wherein pi (i) represents the number of points successfully matched, if pi (i) noteq0 represents p i And q π(i) Is matched, the optimal matching is defined as pi (i) so that the matching cost H (pi) is minimum, and the number of successfully matched points is obtained by using a dynamic programming matching algorithm;
step 2.5, matching the same name plane based on the similarity of the plane shape;
the specific implementation mode of the step 2.5 is as follows;
two-site clouds S and T are known, S is a source point cloud, T is a target point cloud, and the segmented plane set is s= { S 1 ,s 2 ,...,s m Sum t= { T 1 ,t 2 ,...,t n And (2) m, n are the number of planes in the plane set S and the plane set T respectively, and T= { T 1 ,t 2 ,...,t n Plane t in } i And S= { S 1 ,s 2 ,...,s m Plane s in } j The matching cost of the two can be calculated according to the matching cost formula ij And the number N of points successfully matched matched H (pi) and pi (i) in the step 2.4 are respectively corresponding; from the matching result, the similarity of two plane outlinesThe higher their matching cost ij The smaller the size; matching the point cost pcost matched ＝cost ij /N matched Selecting an optimal corresponding plane as a comparison value, and selecting S= { S 1 ,s 2 ,...,s m Intermediate and t i Plane t with minimum matching point cost j Obtain the plane pair (t) i ,s j ) Finally, T= { T is calculated 1 ,t 2 ,...,t n The homonymous planes of all planes in a } set of homonymous planes, Μ= { (t) i ,s i )|t i ∈T；s i ∈S；i＝1,2,...n}；
Step 3, constructing a topological graph according to the matched relationship between the positions of the homonymous planes and the normal vector included angles, and eliminating the mismatching planes based on the homonymous plane edge voting strategy;
and 4, selecting optimal groups of homonymous planes to calculate transformation matrixes of the two-site cloud to finish the registration of the point cloud.
2. A method of point cloud registration using planar shape and topology voting as recited in claim 1, wherein: in the step 1, firstly, a binary coding octree is constructed to complete the voxelization of the point cloud, then, the characteristic of each voxel is calculated, and on the basis, a plane is segmented by adopting a region growing strategy: selecting voxels with obvious plane characteristics as seeds, merging adjacent voxels with similar plane characteristics with the seeds into a cluster, and refining the undivided voxels to improve the segmentation result; finally, accurate plane parameters are obtained by using a RANSAC plane fitting method.
3. A method of point cloud registration using planar shape and topology voting as recited in claim 1, wherein: the specific implementation mode of the step 3 is as follows;
first according to the homonymous plane set Μ= { (t) i ,s i )|t i ∈T；s i E S; i=1, 2,..n } build topology G S (V S ,E S ) And G T (V T ,E T ) The method comprises the steps of carrying out a first treatment on the surface of the Wherein V is S (V T ) Is a node of the topology, each node representing a plane s in M i (t i ) Its attribute is defined by plane s i (t i ) Center point coordinates of (2)Representing that the properties of an edge include two parts: distance of two nodes:and the normal angle between the planes represented by the two nodesWherein->Refers to the coordinates of two nodes, n i ，n j Refers to the normal of the plane;
since M is a set of co-name planes, therefore,for the node correspondence set, if->And->Are correctly matched, then their edge pairs +.>Is also a correct match and in the ideal caseWherein V represents a node set, V represents a node in the set, E represents a set of edges, E represents an element in the set of edges, E (i,j) Representing node i and j connectionsA finished edge; wherein (1)>Represents->Length of->Represents->I.e. the angle between the two planes, if +.>Andthe relation of (2) is represented by->And->At least one of which is a false match;
if it isIs the correct corresponding edge set, consider +.>And->The two corresponding planes with the same name are correctly matched, and then the node set corresponding to the two planes is +.>And->Each 1-vote, this principle is called a plane relation side voting strategy; obviously, a certain node pair ∈ ->Is +.>The more the group of homonymous plane correspondences are, the more robust the group of homonymous plane correspondences are; if the ticket number of a certain node is lower than N ε Then the node is considered to be the wrong homonymous plane match and is taken from Μ = { (t) i ,s i )|t i ∈T；s i E S; i=1, 2,..n } culling, and finally sorting the nodes in M according to the vote count of each node.
4. A method of point cloud registration using planar shape and topology voting as recited in claim 3, wherein: the specific implementation manner of the step 4 is as follows:
selecting three groups of optimal matching planes for calculation, selecting the ground as a first group of homonymous planes according to the constraint condition of maximum area or maximum points or minimum center point Z coordinates in Ping Miandian cloud, and then collecting a homonymous plane set m= { (t) through voting strategies i ,s i )|t i ∈T；s i E S; and selecting two groups of non-parallel planes to form three groups of homonymous planes according to the number of votes by the nodes in i=1, 2, & gt. N } and finally finishing point cloud registration by using a closed form solution estimation rigid body transformation method according to plane parameters of the three groups of homonymous planes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210012942.6A CN114463396B (en) | 2022-01-07 | 2022-01-07 | Point cloud registration method utilizing plane shape and topological graph voting |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210012942.6A CN114463396B (en) | 2022-01-07 | 2022-01-07 | Point cloud registration method utilizing plane shape and topological graph voting |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114463396A CN114463396A (en) | 2022-05-10 |
| CN114463396B true CN114463396B (en) | 2024-02-06 |
Family
ID=81408954
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210012942.6A Active CN114463396B (en) | 2022-01-07 | 2022-01-07 | Point cloud registration method utilizing plane shape and topological graph voting |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114463396B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| 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 |
| CN117934573A (en) * | 2024-03-25 | 2024-04-26 | 北京华航唯实机器人科技股份有限公司 | Point cloud data registration method and device and electronic equipment |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1429264A2 (en) * | 2002-12-05 | 2004-06-16 | Samsung Electronics Co., Ltd. | Method of perceptual 3D shape description and method and apparatus for searching 3D graphics model database using the description method |
| CN109102535A (en) * | 2018-07-27 | 2018-12-28 | 西安理工大学 | A kind of indoor scene method for reconstructing that Component- Based Development extracts |
| CN109523582A (en) * | 2018-09-10 | 2019-03-26 | 中国矿业大学(北京) | A kind of point cloud rough registration method for taking normal vector and multiple dimensioned sparse features into account |
| 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 |
| CN110276790A (en) * | 2019-06-28 | 2019-09-24 | 易思维（杭州）科技有限公司 | Point cloud registration method based on shape constraining |
| CN110443836A (en) * | 2019-06-24 | 2019-11-12 | 中国人民解放军战略支援部队信息工程大学 | A kind of point cloud data autoegistration method and device based on plane characteristic |
| 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 |
| CN111986219A (en) * | 2020-08-10 | 2020-11-24 | 中国科学院光电技术研究所 | Matching method of three-dimensional point cloud and free-form surface model |
| CN112070787A (en) * | 2020-08-10 | 2020-12-11 | 武汉大学 | Aviation three-dimensional point cloud plane segmentation method based on opponent reasoning theory |
| CN112767457A (en) * | 2021-01-25 | 2021-05-07 | 北京影谱科技股份有限公司 | Principal component analysis-based plane point cloud matching method and device |
| CN113516695A (en) * | 2021-05-25 | 2021-10-19 | 中国计量大学 | Point cloud registration strategy in laser profilometer flatness measurement |
| EP3916677A1 (en) * | 2020-05-29 | 2021-12-01 | Faro Technologies, Inc. | Three-dimensional measurement device |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9183631B2 (en) * | 2012-06-29 | 2015-11-10 | Mitsubishi Electric Research Laboratories, Inc. | Method for registering points and planes of 3D data in multiple coordinate systems |
| EP3451291B1 (en) * | 2017-09-01 | 2020-08-26 | Tata Consultancy Services Limited | System and method for shape constrained 3d point cloud registration |
| US20200027266A1 (en) * | 2018-07-17 | 2020-01-23 | Uti Limited Partnership | Building contour generation from point clouds |
-
2022
- 2022-01-07 CN CN202210012942.6A patent/CN114463396B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1429264A2 (en) * | 2002-12-05 | 2004-06-16 | Samsung Electronics Co., Ltd. | Method of perceptual 3D shape description and method and apparatus for searching 3D graphics model database using the description method |
| CN109102535A (en) * | 2018-07-27 | 2018-12-28 | 西安理工大学 | A kind of indoor scene method for reconstructing that Component- Based Development extracts |
| CN109523582A (en) * | 2018-09-10 | 2019-03-26 | 中国矿业大学(北京) | A kind of point cloud rough registration method for taking normal vector and multiple dimensioned sparse features into account |
| 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 |
| CN110443836A (en) * | 2019-06-24 | 2019-11-12 | 中国人民解放军战略支援部队信息工程大学 | A kind of point cloud data autoegistration method and device based on plane characteristic |
| CN110276790A (en) * | 2019-06-28 | 2019-09-24 | 易思维（杭州）科技有限公司 | Point cloud registration method based on shape constraining |
| 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 |
| EP3916677A1 (en) * | 2020-05-29 | 2021-12-01 | Faro Technologies, Inc. | Three-dimensional measurement device |
| CN111986219A (en) * | 2020-08-10 | 2020-11-24 | 中国科学院光电技术研究所 | Matching method of three-dimensional point cloud and free-form surface model |
| CN112070787A (en) * | 2020-08-10 | 2020-12-11 | 武汉大学 | Aviation three-dimensional point cloud plane segmentation method based on opponent reasoning theory |
| CN112767457A (en) * | 2021-01-25 | 2021-05-07 | 北京影谱科技股份有限公司 | Principal component analysis-based plane point cloud matching method and device |
| CN113516695A (en) * | 2021-05-25 | 2021-10-19 | 中国计量大学 | Point cloud registration strategy in laser profilometer flatness measurement |
Non-Patent Citations (6)
| Title |
|---|
| "An Effcient Encoding Voxel-Based Segmentation(EVBS) Algorithm Based on Fast Adjacent Voxel Search for Point Cloud Plane Segmentation";Ming Huang 等;《Remote Sense》;第3页第2节 * |
| Robust point cloud registration based on topological graph and Cauchy weighted lq-norm;Jiayuan Li 等;《ISPRS Journal of Photogrammetry and Remote Sensing》;第244-249页摘要和第3节 * |
| 一种基于几何特征由粗到细点云配准算法;胡加涛;吴晓红;何小海;王正勇;龚剑;;科学技术与工程(第05期);第236-241页 * |
| 基于双目视觉稀疏点云重建的输电线路弧垂测量方法;王礼田;邵凤莹;萧宝瑾;;太原理工大学学报(第06期);第67-71+105页 * |
| 基于数据配准的零件精密装配最佳接触状态研究;张体广 等;《兵工学报》;第39卷(第01期);第127-136页 * |
| 平面特征的多站地面激光雷达点云配准;武鹏;;测绘科学(第09期);第111-114页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114463396A (en) | 2022-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Xia et al. | Geometric primitives in LiDAR point clouds: A review | |
| CN107123164B (en) | Three-dimensional reconstruction method and system for keeping sharp features | |
| CN109410321B (en) | Three-dimensional reconstruction method based on convolutional neural network | |
| CN100559398C (en) | Automatic deepness image registration method | |
| WO2024077812A1 (en) | Single building three-dimensional reconstruction method based on point cloud semantic segmentation and structure fitting | |
| Xu et al. | Reconstruction of scaffolds from a photogrammetric point cloud of construction sites using a novel 3D local feature descriptor | |
| CN107818598B (en) | Three-dimensional point cloud map fusion method based on visual correction | |
| CN114463396B (en) | Point cloud registration method utilizing plane shape and topological graph voting | |
| Tazir et al. | CICP: Cluster Iterative Closest Point for sparse–dense point cloud registration | |
| Yang et al. | Ransac matching: Simultaneous registration and segmentation | |
| Li et al. | A tutorial review on point cloud registrations: principle, classification, comparison, and technology challenges | |
| CN111179321B (en) | Point cloud registration method based on template matching | |
| Yang et al. | Aligning 2.5 D scene fragments with distinctive local geometric features and voting-based correspondences | |
| CN109949350B (en) | Multi-temporal point cloud automatic registration method based on morphological invariant features | |
| Beder et al. | Direct solutions for computing cylinders from minimal sets of 3D points | |
| Wei et al. | Automatic coarse registration of point clouds using plane contour shape descriptor and topological graph voting | |
| CN110111375B (en) | Image matching gross error elimination method and device under Delaunay triangulation network constraint | |
| CN113628263A (en) | Point cloud registration method based on local curvature and neighbor characteristics thereof | |
| Hu et al. | An automatic 3D registration method for rock mass point clouds based on plane detection and polygon matching | |
| CN112396641B (en) | Point cloud global registration method based on congruent two-baseline matching | |
| CN112164145B (en) | Method for rapidly extracting indoor three-dimensional line segment structure based on point cloud data | |
| CN111145129A (en) | Point cloud denoising method based on hyper-voxels | |
| Yue et al. | A new plane segmentation method of point cloud based on mean shift and RANSAC | |
| Guo et al. | Line-based 3d building abstraction and polygonal surface reconstruction from images | |
| CN115082716A (en) | Multi-source point cloud rough matching algorithm for road fine reconstruction |
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 |