CN113256692A - Rigid body registration method and equipment based on adaptive neighborhood weight learning - Google Patents
Rigid body registration method and equipment based on adaptive neighborhood weight learning Download PDFInfo
- Publication number
- CN113256692A CN113256692A CN202110528840.5A CN202110528840A CN113256692A CN 113256692 A CN113256692 A CN 113256692A CN 202110528840 A CN202110528840 A CN 202110528840A CN 113256692 A CN113256692 A CN 113256692A
- Authority
- CN
- China
- Prior art keywords
- point set
- point
- source
- target
- rigid body
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 49
- 230000003044 adaptive effect Effects 0.000 title claims abstract description 27
- 239000011159 matrix material Substances 0.000 claims abstract description 51
- 238000013519 translation Methods 0.000 claims abstract description 28
- 239000013598 vector Substances 0.000 claims abstract description 20
- 230000006870 function Effects 0.000 claims abstract description 15
- 238000012545 processing Methods 0.000 claims abstract description 13
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 8
- 230000009466 transformation Effects 0.000 claims abstract description 7
- 238000004422 calculation algorithm Methods 0.000 claims description 26
- 238000007781 pre-processing Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 230000001174 ascending effect Effects 0.000 claims description 4
- 230000000694 effects Effects 0.000 description 10
- 238000012216 screening Methods 0.000 description 3
- 230000007812 deficiency Effects 0.000 description 2
- 238000012217 deletion Methods 0.000 description 2
- 230000037430 deletion Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000001131 transforming effect Effects 0.000 description 1
Images
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
-
- 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)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
A rigid body registration method and apparatus based on adaptive neighborhood weight learning, the registration method includes carrying on the preconditioning to the source point set and target point set; initializing a rotation matrix and a translation vector, and determining a target function; for each point in the source point set, traversing the target point set to find a corresponding point set with the shortest Euclidean distance; calculating the coordinates of the central positions of the source point set and the corresponding point set, and performing decentralized processing on the two point sets; calculating covariance matrixes of the two point sets after the centralization is removed, performing singular value decomposition, and solving a rotation matrix; obtaining a translation matrix by making a difference between the centralized coordinates of the two point sets; updating the points in the point set through a preset k value; and performing position transformation on the source point set according to the obtained rotation matrix and translation matrix, calculating the sum of squares of corresponding distances of the transformed source point set and the transformed target point set, judging whether a threshold value is reached, outputting a registration result if the threshold value is reached, and returning to iteration if the threshold value is not reached. The invention can improve the registration precision of the point cloud data.
Description
Technical Field
The invention belongs to the field of pose estimation, and relates to a rigid body registration method and equipment based on adaptive neighborhood weight learning.
Background
The main idea of the three-dimensional point cloud registration process is to convert two or more pieces of point cloud data into the same coordinate system through a registration algorithm to obtain a complete three-dimensional model of an object. Currently, point cloud registration methods can be divided into three major categories: a registration method based on geometric features, a registration method based on iteration, a registration method based on curved surfaces, and the like. Three-dimensional point cloud registration is a key research problem of computer vision, and has important application value in the technical fields of pose estimation, reverse engineering, SLAM, image processing and the like at present. The iterative based registration method mainly refers to a classical iterative closest point algorithm, referred to as ICP algorithm for short, proposed by Besl et al. The method is a registration method for solving an optimal solution based on a least square method, and a rotational translation matrix is continuously solved in an iterative mode, so that the distance between two point sets is minimized to realize registration. On the basis of the ICP algorithm, a plurality of improved ICP algorithms are proposed, such as Trim-ICP algorithm, MCC-ICP algorithm and the like. The ICP-based registration algorithm is easily interfered by noise data or data integrity, so that the algorithm is in local optimum, the registration precision is low, and an ideal registration effect cannot be achieved.
The relevant documents are as follows:
[1]Besl P J,McKay N D.Method for registration of 3-D shapes[C]//Sensor fusion IV:control paradigms and data structures.International Society for Optics and Photonics,1992,1611:586-606.
[2]Chetverikov D,Stepanov D,Krsek P.Robust Euclidean alignment of 3D point sets:the trimmed iterative closest point algorithm[J].Image and vision computing,2005,23(3):299-309.
[3]Du S,Xu G,Zhang S,et al.Robust rigid registration algorithm based on pointwise correspondence and correntropy[J].Pattern Recognition Letters,2020,132:91-98.
disclosure of Invention
The invention aims to provide a rigid body registration method and equipment based on adaptive neighborhood weight learning aiming at the problem that the registration effect of the classical ICP algorithm in the prior art on point cloud data containing a large amount of noise data and having deficiency is poor, so as to improve the registration accuracy in the point cloud data containing noise or partial deficiency.
In order to achieve the purpose, the invention has the following technical scheme:
a rigid body registration method based on adaptive neighborhood weight learning comprises the following steps:
preprocessing a source point set P and a target point set;
initializing a rotation matrix R and a translational vector T, initializing a source point set P, and determining a target function;
for each point in the source point set P, traversing the target point set to find a corresponding point set Q with the shortest Euclidean distance;
calculating the central position coordinates of the source point set P and the corresponding point set Q, and performing decentralized processing on the two point sets;
calculating covariance matrixes of the two point sets subjected to the decentralization treatment, and performing singular value decomposition to obtain a rotation matrix R;
obtaining a translation matrix T by making a difference between the centralized coordinates of the two point sets;
updating the points in the point set through a preset k value;
performing position transformation on the source point set P according to the obtained rotation matrix R and translation matrix T, and calculating the square sum d of the corresponding distances of the transformed source point set P and the transformed target point set;
and d is judged whether to reach a preset threshold value, if so, a registration result is output, and if not, iteration is continuously returned.
As a preferred embodiment of the present invention, the preprocessing includes performing sparse processing on a point set with a higher density in the source point set and the target point set.
As a preferred embodiment of the present invention, when initializing the source point set P, a weight is added to each point in the point set, and the determined objective function is as follows:
where w represents the weight vector of the point set, γ represents a penalty parameter for the weight, piAnd q isiAre the corresponding points.
As a preferred scheme of the invention, a KD-Tree algorithm is used for traversing a target point set to find a corresponding point set Q with the nearest Euclidean distance.
As a preferred embodiment of the present invention, the calculation formula for performing the decentralized processing on the two point sets is as follows:
wherein x isiIs the point of the point set P after the ith point is decentralized, yiIs the point of the point set Q after the ith point is decentralized.
As a preferred embodiment of the present invention, the computation formula for computing the covariance matrix of the two point sets after the decentralization processing and performing singular value decomposition is as follows:
S=XYT
U,V=svd(S)
R=Vdiag([1,1,det(VUT)])UT。
as a preferred aspect of the present invention, the calculation expression of the translation matrix T is as follows:
as a preferred embodiment of the present invention, a specific manner of updating the points in the point set is as follows: adding weight to each point in the source point set P, and solving the value according to an objective functionThe values of the points are sorted in ascending order, and the weight vector w corresponding to the point set is updated according to a preset k value;
the calculation process of the update weight vector w is as follows:
the invention also provides a rigid body registration system based on adaptive neighborhood weight learning, which comprises the following steps:
the preprocessing module is used for preprocessing the source point set P and the target point set;
the target function determining module is used for initializing the rotation matrix R and the translation vector T randomly and determining a target function;
the traversal module is used for traversing the target point set for each point in the source point set P to find a corresponding point set Q with the nearest Euclidean distance;
the de-centralization module is used for calculating the central position coordinates of the source point set P and the corresponding point set Q and performing de-centralization processing;
the rotation matrix solving module is used for calculating covariance matrixes of the two point sets after the decentralization treatment and solving a rotation matrix R by carrying out singular value decomposition;
the translation matrix solving module is used for obtaining a translation matrix T by making a difference between the centralized coordinates of the two point sets;
the point set updating module is used for updating the points in the point set through a preset k value;
the point set distance square sum solving module is used for carrying out position transformation on the source point set P according to the rotation matrix R and the translation matrix T and calculating the corresponding distance square sum d of the transformed source point set P and the transformed target point set;
and the registration output module is used for judging whether the sum of squares d of the corresponding distances of the source point set P and the target point set reaches a preset threshold value, outputting a registration result if the sum of squares d of the corresponding distances of the source point set P and the target point set reaches the preset threshold value, and continuing to return to iteration if the sum of squares d of the corresponding distances of the source point set P and the target point set does not reach the preset threshold value.
The invention also provides a computer readable storage medium storing a computer program which, when executed by a processor, implements the rigid body registration method based on adaptive neighborhood weight learning.
Compared with the prior art, the invention has the following beneficial effects:
the corresponding point weights in the point set are iteratively updated by setting k self-adaptive neighborhoods, so that the robustness of the point cloud data containing a large amount of noise is strong. By giving a larger value to the weight of a point closer to the noise point and screening or giving a smaller weight to a point farther away from the noise point, the noise point can be effectively screened and excluded. By setting a proper k value, the effect of screening out all noises can be achieved, so that the precision of the registration algorithm is greatly improved. In addition, compared with the conventional ICP algorithm, the method has great advantages on the point cloud data with local deletion, and the registration accuracy is greatly improved.
Drawings
FIG. 1 is a flowchart of a rigid body registration method based on adaptive neighborhood weight learning according to an embodiment of the present invention;
FIG. 2 is a flowchart of a method for updating adaptive neighborhood weights according to an embodiment of the present invention;
fig. 3 is a graph comparing the effect of the present invention and the conventional ICP algorithm on 2D data:
(a) a conventional ICP; (b) Trim-ICP; (c) MCC-ICP; (d) the method of the invention;
FIG. 4 is a graph comparing the effect of the present invention and previous ICP algorithm on 3D model data:
(a) a conventional ICP; (b) Trim-ICP; (c) MCC-ICP; (d) the method of the invention;
fig. 5 is a comparison graph of the effect of the ICP algorithm of the present invention on 3D synthetic scene data:
(a) a conventional ICP; (b) Trim-ICP; (c) MCC-ICP; (d) the method of the invention;
fig. 6 is a comparison graph of the effect of the ICP algorithm of the present invention on 3D real scene data:
(a) a conventional ICP; (b) Trim-ICP; (c) MCC-ICP; (d) the method of the invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Referring to fig. 1, the rigid body registration method based on adaptive neighborhood weight learning provided by the invention comprises the following steps:
s1, preprocessing the source point set and the target point set, and performing down-sampling and other steps;
s2, randomly initializing a rotation matrix R and a translational vector T, and determining a target function of the algorithm;
s3, traversing each point in the target point set to find a corresponding point set Q with the closest distance for each point in the source point set P;
s4, calculating the coordinates of the center positions of the point set P and the corresponding point set Q, and meanwhile, performing decentralization on the point set P and the point set Q;
s5, calculating covariance matrixes of the two point sets after the centralization is removed, and decomposing to obtain a rotation matrix R;
s6, calculating the translation vector T by the difference of the centers of the two point sets;
s7, updating the weight corresponding to each point set according to the preset k value;
s8, transforming the point set P, and calculating the square sum d of the corresponding distance between the transformed point set P and the target point set Q;
and S9, when the sum of squared distances d is smaller than a preset threshold, stopping iteration by the algorithm, otherwise, repeating the steps S3 to S8 until the stop condition is met, ending iteration, and outputting a registration result.
Step S1 is to pre-process the point cloud data, and mainly for the point set with a large density, it needs to first perform sparse processing, i.e. down-sampling of the point cloud, and the down-sampled data can represent the shape of the original point set and can also speed up the algorithm.
Step S2 is to perform initial change on the source point set P according to the initialized rotation matrix R and translation matrix T, and further obtain an algorithm objective function, where an expression of the objective function is as follows:
where w represents the weight vector of the point set, γ represents a penalty parameter for the weight, piAnd q isiAre the corresponding points.
And step S3, searching through the kd-tree to quicken finding the corresponding point with the closest distance of the source click corresponding target point set.
Step S4 decentralizes the point set P and the point set Q, and the calculation formula is as follows:
wherein x isiIs the point of the point set P after the ith point is decentralized, yiIs the point of the point set Q after the ith point is decentralized.
Step S5 calculates the covariance matrices of the two point sets after decentralization, and decomposes the covariance matrices to obtain a rotation matrix R, where the calculation formula is as follows:
S=XYT
U,V=svd(S)
R=Vdiag([1,1,det(VUT)])UT
step S6 is to take the difference between the two point set centralized coordinates to obtain a translation matrix T, and the calculation formula is as follows:
step S7 is rightAnd (5) performing ascending sorting, and updating the weight vector w corresponding to the point set according to a preset k value. Where k represents k pre-set adaptive neighborhoods.
Step S8 performs position transformation on the point set P according to the obtained rotation matrix R and translation matrix T, and obtains the distance sum of squares d of the transformed point set P from the point set Q, where the calculation formula is as follows:
if the distance error d is smaller than the threshold set by the algorithm in step S9, the iteration is stopped. Otherwise, the iteration is continued.
Referring to fig. 2, the adaptive neighborhood weight updating method includes the following steps:
S12, sorting the distance vectors in ascending order as follows:
and S13, updating the weight corresponding to the point set according to the preset k value. The update formula is as follows:
table 1 shows that the root mean square MSE of the distance between two point sets containing noise, the root mean square error of the rotation matrix R and the translation vector T are used as final indexes to evaluate the selected data.
TABLE 1
Methods | MSE | R_error | T_error |
Classical ICP method | 6.1×10-3 | 470×10-3 | 16.5×10-3 |
Trim-ICP method | 4.1×10-3 | 141×10-3 | 10.5×10-3 |
MCC-ICP method | 2.3×10-3 | 5×10-3 | 1.4×10-4 |
The method of the invention | 1.4×10-3 | 1.6×10-3 | 7.09×10-5 |
As can be seen from Table 1, the method provided by the invention has remarkable effects of three indexes in the registration process of the point cloud containing noise.
Table 2 shows that the present invention uses the root mean square MSE, the rotation matrix R, and the root mean square error of the translation vector T, which include the distances of the partially overlapped point sets, as final indicators to evaluate on the selected data.
TABLE 2
As can be seen from Table 2, the effect of the three indexes in the registration process of the point cloud containing partial overlap is very significant.
The ICP algorithm based on the adaptive neighborhood weight learning sets the weights of the corresponding points in the k adaptive neighborhoods and the point set to be iteratively updated, and has strong robustness on point cloud data containing a large amount of noise. The weight of the points with the closer distance is given a larger value, and the points with the farther distance are screened out or given a smaller weight, so that the noise points are effectively screened out and eliminated. By setting a proper k value, the effect of screening out all noises can be achieved, and the precision of the registration algorithm is greatly improved. In addition, compared with the conventional ICP algorithm, the method has great advantages on the point cloud data with local deletion, and the registration accuracy is greatly improved.
The invention also provides a rigid body registration system based on adaptive neighborhood weight learning, which comprises the following steps:
the preprocessing module is used for preprocessing the source point set and the target point set;
the target function determining module is used for initializing the rotation matrix R and the translation vector T randomly and determining a target function;
the traversal module is used for traversing the target point set for each point in the source point set P to find a corresponding point set Q with the nearest Euclidean distance;
the de-centralization module is used for calculating the central position coordinates of the source point set P and the corresponding point set Q and performing de-centralization processing;
the rotation matrix solving module is used for calculating covariance matrixes of the two point sets after the decentralization treatment and solving a rotation matrix R by carrying out singular value decomposition;
the translation matrix solving module is used for obtaining a translation matrix T by making a difference between the centralized coordinates of the two point sets;
the point set updating module is used for updating the points in the point set through a preset k value;
the point set distance square sum solving module is used for carrying out position transformation on the source point set P according to the rotation matrix R and the translation matrix T and calculating the corresponding distance square sum d of the transformed source point set P and the transformed target point set Q;
and the registration output module is used for judging whether the sum of squares d of the corresponding distances of the source point set P and the target point set Q reaches a preset threshold value, outputting a registration result if the sum of squares d of the corresponding distances of the source point set P and the target point set Q reaches the preset threshold value, and continuing to return to iteration if the sum of squares d of the corresponding distances of the source point set P and the target point set Q does not reach the preset threshold value.
The invention also provides a computer readable storage medium storing a computer program which, when executed by a processor, implements the rigid body registration method based on adaptive neighborhood weight learning.
The computer program may be partitioned into one or more modules/units stored in the memory and executed by the processor to perform the rigid body registration method of the present invention.
The processor may be a Central Processing Unit (CPU), other general purpose processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The memory may be used to store computer programs and/or modules, and the processor implements the various functions of the rigid body registration system of the present invention by executing or executing the computer programs and/or modules stored in the memory, as well as by recalling data stored in the memory.
The above-mentioned embodiments are only preferred embodiments of the present invention, and are not intended to limit the technical solution of the present invention, and it should be understood by those skilled in the art that the technical solution can be modified and replaced by a plurality of simple modifications and replacements without departing from the spirit and principle of the present invention, and the modifications and replacements also fall into the protection scope covered by the claims.
Claims (10)
1. A rigid body registration method based on adaptive neighborhood weight learning is characterized by comprising the following steps:
preprocessing a source point set P and a target point set;
initializing a rotation matrix R and a translational vector T, initializing a source point set P, and determining a target function;
for each point in the source point set P, traversing the target point set to find a corresponding point set Q with the shortest Euclidean distance;
calculating the central position coordinates of the source point set P and the corresponding point set Q, and performing decentralized processing on the two point sets;
calculating covariance matrixes of the two point sets subjected to the decentralization treatment, and performing singular value decomposition to obtain a rotation matrix R;
obtaining a translation matrix T by making a difference between the centralized coordinates of the two point sets;
updating the points in the point set through a preset k value;
performing position transformation on the source point set P according to the obtained rotation matrix R and translation matrix T, and calculating the square sum d of the corresponding distances of the transformed source point set P and the transformed target point set;
and d is judged whether to reach a preset threshold value, if so, a registration result is output, and if not, iteration is continuously returned.
2. The rigid body registration method based on adaptive neighborhood weight learning of claim 1, wherein: the preprocessing comprises the step of carrying out sparse processing on the point set with higher density in the source point set and the target point set.
3. The rigid body registration method based on adaptive neighborhood weight learning according to claim 1, wherein when initializing the source point set P, adding weight to each point in the point set, and determining the objective function as follows:
where w represents the weight vector of the point set, γ represents a penalty parameter for the weight, piAnd q isiAre the corresponding points.
4. The rigid body registration method based on adaptive neighborhood weight learning of claim 1, wherein: and traversing the target point set by using a KD-Tree algorithm to find the corresponding point set Q with the shortest Euclidean distance.
5. The rigid body registration method based on adaptive neighborhood weight learning according to claim 1, wherein the calculation formula for de-centering the two point sets is as follows:
wherein x isiIs the point of the point set P after the ith point is decentralized, yiIs the point of the point set Q after the ith point is decentralized.
6. The rigid body registration method based on adaptive neighborhood weight learning of claim 1, wherein the covariance matrix of the two de-centered point sets is calculated, and singular value decomposition is performed according to the following formula:
S=XYT
U,V=svd(S)
R=Vdiag([1,1,det(VUT)])UT。
8. the rigid body registration method based on adaptive neighborhood weight learning according to claim 1, wherein the specific way to update the points in the point set is as follows: adding weight to each point in the source point set P, and solving the value according to an objective functionThe values of the points are sorted in ascending order, and the weight vector w corresponding to the point set is updated according to a preset k value;
the calculation process of the update weight vector w is as follows:
9. a rigid body registration system based on adaptive neighborhood weight learning, comprising:
the preprocessing module is used for preprocessing the source point set P and the target point set;
the target function determining module is used for initializing the rotation matrix R and the translation vector T randomly and determining a target function;
the traversal module is used for traversing the target point set for each point in the source point set P to find a corresponding point set Q with the nearest Euclidean distance;
the de-centralization module is used for calculating the central position coordinates of the source point set P and the corresponding point set Q and performing de-centralization processing;
the rotation matrix solving module is used for calculating covariance matrixes of the two point sets after the decentralization treatment and solving a rotation matrix R by carrying out singular value decomposition;
the translation matrix solving module is used for obtaining a translation matrix T by making a difference between the centralized coordinates of the two point sets;
the point set updating module is used for updating the points in the point set through a preset k value;
the point set distance square sum solving module is used for carrying out position transformation on the source point set P according to the rotation matrix R and the translation matrix T and calculating the corresponding distance square sum d of the transformed source point set P and the transformed target point set;
and the registration output module is used for judging whether the sum of squares d of the corresponding distances of the source point set P and the target point set reaches a preset threshold value, outputting a registration result if the sum of squares d of the corresponding distances of the source point set P and the target point set reaches the preset threshold value, and continuing to return to iteration if the sum of squares d of the corresponding distances of the source point set P and the target point set does not reach the preset threshold value.
10. A computer-readable storage medium storing a computer program, characterized in that: the computer program when executed by a processor implements the rigid body registration method based on adaptive neighborhood weight learning of any of claims 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110528840.5A CN113256692B (en) | 2021-05-14 | 2021-05-14 | Rigid body registration method and device based on self-adaptive neighborhood weight learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110528840.5A CN113256692B (en) | 2021-05-14 | 2021-05-14 | Rigid body registration method and device based on self-adaptive neighborhood weight learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113256692A true CN113256692A (en) | 2021-08-13 |
CN113256692B CN113256692B (en) | 2024-04-02 |
Family
ID=77181972
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110528840.5A Active CN113256692B (en) | 2021-05-14 | 2021-05-14 | Rigid body registration method and device based on self-adaptive neighborhood weight learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113256692B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116523979A (en) * | 2023-04-24 | 2023-08-01 | 北京长木谷医疗科技股份有限公司 | Point cloud registration method and device based on deep learning and electronic equipment |
CN116563356A (en) * | 2023-05-12 | 2023-08-08 | 北京长木谷医疗科技股份有限公司 | Global 3D registration method and device and electronic equipment |
Family Cites Families (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN105551015A (en) * | 2015-12-02 | 2016-05-04 | 南京邮电大学 | Scattered-point cloud image registering method |
CN110634161B (en) * | 2019-08-30 | 2023-05-05 | 哈尔滨工业大学(深圳) | Rapid high-precision estimation method and device for workpiece pose based on point cloud data |
CN110866969B (en) * | 2019-10-18 | 2022-06-14 | 西北工业大学 | Engine blade reconstruction method based on neural network and point cloud registration |
CN112529945B (en) * | 2020-11-17 | 2023-02-21 | 西安电子科技大学 | Multi-view three-dimensional ISAR scattering point set registration method |
-
2021
- 2021-05-14 CN CN202110528840.5A patent/CN113256692B/en active Active
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN116523979A (en) * | 2023-04-24 | 2023-08-01 | 北京长木谷医疗科技股份有限公司 | Point cloud registration method and device based on deep learning and electronic equipment |
CN116523979B (en) * | 2023-04-24 | 2024-01-30 | 北京长木谷医疗科技股份有限公司 | Point cloud registration method and device based on deep learning and electronic equipment |
CN116563356A (en) * | 2023-05-12 | 2023-08-08 | 北京长木谷医疗科技股份有限公司 | Global 3D registration method and device and electronic equipment |
CN116563356B (en) * | 2023-05-12 | 2024-06-11 | 北京长木谷医疗科技股份有限公司 | Global 3D registration method and device and electronic equipment |
Also Published As
Publication number | Publication date |
---|---|
CN113256692B (en) | 2024-04-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111897349B (en) | Autonomous obstacle avoidance method for underwater robot based on binocular vision | |
CN113256692A (en) | Rigid body registration method and equipment based on adaptive neighborhood weight learning | |
CN112017220A (en) | Point cloud accurate registration method based on robust constraint least square algorithm | |
CN113178009A (en) | Indoor three-dimensional reconstruction method utilizing point cloud segmentation and grid repair | |
CN112651944B (en) | 3C component high-precision six-dimensional pose estimation method and system based on CAD model | |
CN102129716A (en) | Method for fitting curved surface of water-fire bending plate | |
CN112801945A (en) | Depth Gaussian mixture model skull registration method based on dual attention mechanism feature extraction | |
CN115359083A (en) | Method, system, medium, and electronic device for obtaining contour of obstacle | |
CN117132630A (en) | Point cloud registration method based on second-order spatial compatibility measurement | |
Drwięga | Features matching based merging of 3D maps in multi-robot systems | |
CN115661218B (en) | Virtual super-point-based laser point cloud registration method and system | |
Chan et al. | A 3D-point-cloud feature for human-pose estimation | |
CN116758212A (en) | 3D reconstruction method, device, equipment and medium based on self-adaptive denoising algorithm | |
CN115601408A (en) | Point cloud registration method based on particle swarm optimization and topological graph | |
Bharati et al. | Res-q: Robust outlier detection algorithm for fundamental matrix estimation | |
Drwięga | Incremental 3d maps server based on feature matching for multi-robot systems | |
Patoommakesorn et al. | The 3d edge reconstruction from 2d image by using correlation based algorithm | |
Wong et al. | 3d reconstruction from multiple views using particle swarm optimization | |
CN112767462A (en) | Point cloud single point alignment method based on ridge-valley feature and depth feature descriptors | |
CN112116653A (en) | Object posture estimation method for multiple RGB pictures | |
Xiong et al. | Machine vision for 3D mechanical part recognition in intelligent manufacturing environments | |
Lee et al. | An improved stereo matching algorithm with robustness to noise based on adaptive support weight | |
Mondal et al. | Performance review of the stereo matching algorithms | |
CN116797803B (en) | Point cloud matching method based on sectional decoupling, electronic equipment and medium | |
Song et al. | Evolutionary Pose Measurement by Stereo Model Matching. |
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 |