CN113139991A - 3D point cloud registration method based on overlapping region mask prediction - Google Patents
3D point cloud registration method based on overlapping region mask prediction Download PDFInfo
- Publication number
- CN113139991A CN113139991A CN202110521230.2A CN202110521230A CN113139991A CN 113139991 A CN113139991 A CN 113139991A CN 202110521230 A CN202110521230 A CN 202110521230A CN 113139991 A CN113139991 A CN 113139991A
- Authority
- CN
- China
- Prior art keywords
- point cloud
- features
- source
- overlapping area
- target point
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 36
- 230000009466 transformation Effects 0.000 claims abstract description 21
- 230000008569 process Effects 0.000 claims abstract description 8
- 238000004364 calculation method Methods 0.000 claims description 16
- 230000006870 function Effects 0.000 claims description 12
- 230000009467 reduction Effects 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 7
- 241000282326 Felis catus Species 0.000 claims description 6
- 238000010606 normalization Methods 0.000 claims description 3
- 230000008447 perception Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 abstract description 5
- 238000013527 convolutional neural network Methods 0.000 abstract 1
- 230000009286 beneficial effect Effects 0.000 description 4
- 230000003190 augmentative effect Effects 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 238000013507 mapping Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000004807 localization Effects 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000001360 synchronised 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
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
- G06T7/344—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods involving models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a 3D point cloud registration method based on overlapping region mask prediction, which uses an iterative convolutional neural network to learn point cloud rigid body transformation from a coarse mode to a fine mode. The 3D rigid body transformation is characterized by quaternions, which can avoid the situation of a gimbaled lock, and translation distances along the directions of the three XYZ axes. Furthermore, we learn a mask with attention mechanism in the network. The mask not only highlights the overlapping parts of the source point cloud and the target point cloud, but also can filter out the non-overlapping areas and the interference of noise. After each iteration process, when the mask predicted by the iteration is used in the point cloud feature extraction process of the next iteration, the prediction of the point cloud overlapping region and the regression of the 3D rigid body transformation are guaranteed to be mutually promoted.
Description
Technical Field
The invention relates to the field of computational graphics and computer vision, in particular to a 3D point cloud registration method based on overlapping region mask prediction.
Background
The 3D point cloud registration is a process of matching And superimposing two or more point clouds acquired at different times, at different viewing angles And under different sensors, And this technique is widely applied to applications And fields such as three-dimensional Scene Reconstruction (3D Scene Reconstruction), synchronous positioning And Mapping (Simultaneous Localization And Mapping), Augmented Reality (Augmented Reality), And automatic driving (Autopilot).
Among the existing various Point cloud registration methods, an algorithm based on nearest neighbor iteration (Iterative Closest Point) is widely applied due to its simplicity and efficiency. The method comprises the steps of repeatedly calculating nearest neighbor points of coordinates in an Euclidean space in an iteration mode to determine matching point pairs, solving a rigid body transformation matrix in a Singular Value Decomposition (Singular Value Decomposition) mode, wherein the solving precision highly depends on the difference size, the noise size and the overlapping degree of the initial positions of a source point cloud and a target point cloud. When scenes with large initial position difference, strong noise interference and small overlapping degree are processed, the matching point pairs determined by the method usually have errors, so that the scenes cannot be normally registered. Researchers have subsequently proposed DNN-based methods to learn robust depth features that can successfully handle large initial position differences and strong noise scenes. However, when the overlapping area of the source point cloud and the target point cloud is small, the mismatch of the corresponding points is easily caused. In order to enable the algorithm to have stronger registration capability for partially overlapped point clouds, a method based on external point filtering is provided, but the registration accuracy is difficult to improve because corresponding point pairs still need to be matched and only information of sparse partial points in the point clouds can be utilized.
Disclosure of Invention
In order to overcome the defects in the prior art, the invention provides a 3D point cloud registration method based on overlapping region mask prediction.
In order to achieve the purpose of the invention, the invention adopts the technical scheme that:
A3D point cloud registration method based on overlapping region mask prediction comprises the following steps:
the method comprises the following steps:
s1, extracting the features of the source point cloud and the target point cloud to obtain the features of each point;
s2, performing multilayer convolution operation on the features of each point obtained in the step S1 to obtain global features of the source point cloud and the target point cloud at the same latitude, and calculating the mutual overlapping area of the source point cloud and the target point cloud;
and S3, combining the features of each point obtained in the step S1 and the global features obtained in the step S2, calculating the 3D rigid body transformation parameters, and iterating.
The method has the advantages that the information of all points in the point cloud can be utilized, meanwhile, the classification result of the point cloud overlapping region can be utilized to avoid the influence of the characteristics of non-overlapping partial points on the global characteristics of the point cloud, and then the regression of rigid body transformation parameters is promoted to be more accurate.
Further, the step S1 specifically includes:
s11, extracting the characteristics of each point in the source point cloud and the target point cloud;
s12, weighting the characteristics obtained in the step S11 by using a mask of iterative prediction;
s13, performing maximum pooling operation on the characteristics of the points weighted in the step S12 to obtain respective global characteristics of the source point cloud and the target point cloud;
and S13, copying a plurality of global features obtained in the step S13, and combining the copied global features with the intermediate layer features of the points in the source point cloud and the target point cloud respectively to obtain the features of each point in the source point cloud and the target point cloud.
The method has the advantages that information interaction between the source point cloud and the target point cloud can be carried out, the subsequent point cloud overlapping area classification network is endowed with the capability of distinguishing whether the points are positioned in the overlapping area, and accurate classification of the subsequent point cloud overlapping area is facilitated.
Further, the step S2 specifically includes:
the step S2 specifically includes:
s21, performing multilayer convolution calculation on the features of each point obtained in the step S1 to obtain point cloud features after dimension reduction, and extracting the intermediate layer features after dimension reduction each time;
s22, calculating the point cloud characteristics after dimensionality reduction through a Softmax function to obtain the probability that each point belongs to an overlapping area, and calculating through an argmax function to obtain a classification result;
and S23, obtaining an overlapping area of the source point cloud and the target point cloud according to the classification result of the step S22.
Further, the calculation manner of obtaining the classification result through the argmax function in step S22 is as follows:
wherein y represents the classification result of the kth point in the point cloud, z represents the confidence value of the network prediction, C represents two categories of belonging to an overlapping area and not belonging to the overlapping area, and N represents the number of coordinate points in the point cloud.
The method has the advantages that the overall characteristics and the characteristics of each point in the source point cloud and the target point cloud can be comprehensively utilized, so that whether each point in the point cloud belongs to the overlapping area or not can be accurately judged, and finally, the interference of the characteristics of the non-overlapping part points on the registration process is avoided.
Further, the step S3 specifically includes:
the step S3 specifically includes:
s31, combining the characteristics of each point in the source point cloud and the target point cloud obtained in the step S1 and the characteristics of the middle layer obtained in the step S21;
s32, performing maximum pooling calculation on the combined features obtained in the step S31 to obtain the overall features of the source point cloud and the target point cloud;
s33, performing multilayer perception regression calculation on the overall characteristics obtained in the step S32 to obtain quaternion and translation distance, and performing normalization processing on the quaternion to obtain 3D rigid body transformation after one iteration;
and S34, applying the 3D rigid body transformation obtained in the step S33 to the source point cloud, and repeating the steps S31-S33 on the transformed source point cloud to obtain the 3D target point cloud after registration.
Further, the manner of calculating the quaternion and the translation distance by the multilayer perceptual regression calculation in step S33 is as follows:
{q，t}＝rθ(cat[fx，fY])
wherein q representsQuaternion and t denote the translation distance, rθ() Representing a multilayer perceptual regression network, cat]Indicating a cascade operation, fXAnd fYGlobal features representing the source point cloud and the target point cloud.
The technical scheme has the advantages that the intermediate layer features and the global features during point cloud feature extraction are fused, feature information for registration is enriched, the feature learning difficulty of the whole registration network is reduced in the iterative registration process, the network in different iterations is benefited to pay attention to rigid body transformation information in different distributions, and the registration accuracy is improved finally.
Further, the loss function of the overlapping region of the source point cloud and the target point cloud calculated in step S2 is expressed as:
wherein g and p respectively represent the label of the point cloud overlapping area and the probability of the point cloud overlapping area predicted by the network, i represents the ith iteration process, alpha represents the proportion of the overlapping area of the source point cloud and the target point cloud, and M represents the overlapping area mask.
The beneficial effect of the scheme is that the classification loss weighted by the proportion coefficient is beneficial to the contribution of positive and negative samples in the balancing process, and further the classification accuracy is improved.
Further, the regression loss of the rigid body transformation in step S33 is represented as:
where g denotes the true label and λ denotes the coefficient that balances the two loss terms.
The beneficial effect of the above scheme is that the quaternion is directly constrained to avoid the occurrence of the universal lock condition, and meanwhile, the proportional coefficient is added to balance the two losses, which is beneficial for the network to uniformly optimize the quaternion and the translation distance.
Drawings
Fig. 1 is a schematic flow chart of a 3D point cloud registration method based on overlapping region mask prediction according to the present invention.
Detailed Description
The following description of the embodiments of the present invention is provided to facilitate the understanding of the present invention by those skilled in the art, but it should be understood that the present invention is not limited to the scope of the embodiments, and it will be apparent to those skilled in the art that various changes may be made without departing from the spirit and scope of the invention as defined and defined in the appended claims, and all matters produced by the invention using the inventive concept are protected.
A3D point cloud registration method based on overlapping region mask prediction comprises the following steps:
s1, extracting the features of the source point cloud and the target point cloud to obtain the features of each point;
in particular to
And S11, extracting the features of each point in the source point cloud and the target point cloud, wherein in the embodiment, the point cloud pair is input into a Feature extraction network (Feature Extractor), and Feature extraction is firstly carried out on each point in the point cloud.
S12, weighting the characteristics obtained in the step S11 by using a mask of iterative prediction;
s13, performing maximum pooling operation on the characteristics of the points weighted in the step S12 to obtain respective global characteristics of the source point cloud and the target point cloud; weighting the features by using the mask of the last iteration prediction, and then obtaining the respective global features of the source point cloud and the target point cloud through maximum pooling.
And S13, copying and duplicating the global features obtained in the step S13, combining the copied global features with the intermediate layer features of the points in the source point cloud and the target point cloud respectively, and fusing to obtain the features of each point containing the information of the source point cloud and the target point cloud.
S2, performing multilayer convolution operation on the features of each point obtained in the step S1 to obtain global features of the source point cloud and the target point cloud at the same latitude, and calculating the mutual overlapping area of the source point cloud and the target point cloud;
and (3) sending the characteristics of each point into a Mask Predictor, and predicting the overlapped area of the source point cloud and the target point cloud, wherein the overlapped area is represented as 1, and the non-overlapped area is represented as 0.
The specific operation is as follows:
s21, performing multilayer convolution calculation on the features of each point obtained in the step S1 to obtain point cloud features after dimension reduction, and extracting the intermediate layer features after dimension reduction each time; and after the feature of each point is subjected to the multilayer convolution layer, continuously reducing the dimension of the feature, and finally obtaining the feature of each point with the dimension of 2.
S22, calculating the point cloud characteristics after dimensionality reduction through a Softmax function to obtain the probability that each point belongs to an overlapping area, and calculating through an argmax function to obtain a classification result, wherein the specific calculation mode is as follows:
wherein y represents the classification result of the kth point in the point cloud, z represents the confidence value of the network prediction, C represents two categories of belonging to an overlapping area and not belonging to the overlapping area, and N represents the number of coordinate points in the point cloud.
The predicted mask will be applied to two places: the method comprises the steps of weighting the point cloud features during extraction and weighting the intermediate layer features predicted by the mask during regression of 3D rigid body transformation parameters to shield interference of outliers.
And S23, obtaining an overlapping area of the source point cloud and the target point cloud according to the classification result of the step S22.
And S3, combining the characteristics of each point obtained in the step S1 and the characteristics of the middle layer obtained in the step S2, calculating the 3D rigid body transformation parameters, and iterating.
In this embodiment, the step S3 specifically includes:
s31, combining the characteristics of each point in the source point cloud and the target point cloud obtained in the step S1 and the characteristics of the middle layer obtained in the step S21; in this embodiment, the characteristics of each point of the source point cloud and the target point cloud and the characteristics of the intermediate layer of the mask prediction module are combined and then sent to a 3D rigid body Transformation parameter Regression module (Transformation Regression).
S32, performing maximum pooling calculation on the combined features obtained in the step S31 to obtain the overall features of the source point cloud and the target point cloud,
s33, performing multilayer perception regression calculation on the overall characteristics obtained in the step S32 to obtain quaternion and translation distance, and performing normalization processing on the quaternion to obtain 3D rigid body transformation after one iteration; the method for obtaining the quaternion and the translation distance through multilayer perceptual regression calculation comprises the following steps:
(q,t)＝rθ(cat[fx，fY])
where q denotes a quaternion and t denotes a translation distance, rθ() Representing a multilayer perceptual regression network, cat]Indicating a cascade operation, fXAnd fYGlobal features representing the source point cloud and the target point cloud.
And S34, applying the 3D rigid body transformation obtained in the step S33 to the source point cloud, and repeating the steps S31-S33 on the transformed source point cloud to obtain the 3D target point cloud after registration.
Further, the loss function of the overlapping region of the source point cloud and the target point cloud calculated in step S2 is expressed as:
g and p respectively represent a label of a point cloud overlapping area and the probability of the point cloud overlapping area predicted by the network, represent the ith iteration process, alpha represents the proportion of the overlapping area of the source point cloud and the target point cloud, and M represents an overlapping area mask.
Further, the regression loss of the rigid body transformation in step S33 is represented as:
where i denotes the ith iteration, g denotes the true label, and λ denotes the coefficient that balances the two loss terms. The loss of constrained quaternion uses the L1 distance and the loss of constrained translation distance uses the L2 distance. This results in the final total loss function, which is the sum of the two losses in each iteration as follows:
it will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.
Claims (8)
1. A3D point cloud registration method based on overlapping region mask prediction is characterized by comprising the following steps:
s1, extracting the features of the source point cloud and the target point cloud to obtain the features of each point;
s2, performing multilayer convolution operation on the features of each point obtained in the step S1 to obtain global features of the source point cloud and the target point cloud at the same latitude, and calculating the mutual overlapping area of the source point cloud and the target point cloud;
and S3, combining the features of each point obtained in the step S1 and the global features obtained in the step S2, calculating the 3D rigid body transformation parameters, and iterating.
2. The 3D point cloud registration method based on overlap region mask prediction according to claim 1, wherein the step S1 specifically comprises:
s11, extracting the characteristics of each point in the source point cloud and the target point cloud;
s12, weighting the characteristics obtained in the step S11 by using a mask of iterative prediction;
s13, performing maximum pooling operation on the characteristics of the points weighted in the step S12 to obtain respective global characteristics of the source point cloud and the target point cloud;
and S13, copying a plurality of global features obtained in the step S13, and combining the copied global features with the intermediate layer features of the points in the source point cloud and the target point cloud respectively to obtain the features of each point in the source point cloud and the target point cloud.
3. The 3D point cloud registration method based on overlap region mask prediction according to claim 2, wherein the step S2 specifically includes:
s21, performing multilayer convolution calculation on the features of each point obtained in the step S1 to obtain point cloud features after dimension reduction, and extracting the intermediate layer features after dimension reduction each time;
s22, calculating the point cloud characteristics after dimensionality reduction through a Softmax function to obtain the probability that each point belongs to an overlapping area, and calculating through an argmax function to obtain a classification result;
and S23, obtaining an overlapping area of the source point cloud and the target point cloud according to the classification result of the step S22.
4. The 3D point cloud registration method based on overlap region mask prediction according to claim 3, wherein the calculation manner of the classification result obtained by the argmax function calculation in step S22 is as follows:
wherein y represents the classification result of the kth point in the point cloud, z represents the confidence value of the network prediction, C represents two categories of belonging to an overlapping area and not belonging to the overlapping area, and N represents the number of coordinate points in the point cloud.
5. The 3D point cloud registration method based on overlap region mask prediction according to claim 4, wherein the step S3 specifically comprises:
s31, combining the characteristics of each point in the source point cloud and the target point cloud obtained in the step S1 and the characteristics of the middle layer obtained in the step S21;
s32, performing maximum pooling calculation on the combined features obtained in the step S31 to obtain the overall features of the source point cloud and the target point cloud;
s33, performing multilayer perception regression calculation on the overall characteristics obtained in the step S32 to obtain quaternion and translation distance, and performing normalization processing on the quaternion to obtain 3D rigid body transformation after one iteration;
and S34, applying the 3D rigid body transformation obtained in the step S33 to the source point cloud, and repeating the steps S31-S33 on the transformed source point cloud to obtain the 3D target point cloud after registration.
6. The 3D point cloud registration method based on overlap region mask prediction of claim 5, wherein the way of calculating the quaternion and the translation distance by multi-layer perceptual regression calculation in the step S33 is as follows:
{q，t}＝rθ(cat[fX，fY])
where q denotes a quaternion and t denotes a translation distance, rθ() Representing a multilayer perceptual regression network, cat]Indicating a cascade operation, fXAnd fYGlobal features representing the source point cloud and the target point cloud.
7. The overlap region mask prediction-based 3D point cloud registration method according to claim 6, wherein the loss function of the mutual overlap region of the source point cloud and the target point cloud calculated in the step S2 is represented as:
wherein g and p respectively represent the label of the point cloud overlapping area and the probability of the point cloud overlapping area predicted by the network, i represents the ith iteration process, alpha represents the proportion of the overlapping area of the source point cloud and the target point cloud, and M represents the overlapping area mask.
8. The overlap region mask prediction-based 3D point cloud registration method of claim 7, wherein the regression loss of the rigid body transformation in the step S33 is represented as:
where g denotes the true label and λ denotes the coefficient that balances the two loss terms.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110521230.2A CN113139991A (en) | 2021-05-13 | 2021-05-13 | 3D point cloud registration method based on overlapping region mask prediction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110521230.2A CN113139991A (en) | 2021-05-13 | 2021-05-13 | 3D point cloud registration method based on overlapping region mask prediction |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113139991A true CN113139991A (en) | 2021-07-20 |
Family
ID=76817861
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110521230.2A Pending CN113139991A (en) | 2021-05-13 | 2021-05-13 | 3D point cloud registration method based on overlapping region mask prediction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113139991A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114004871A (en) * | 2022-01-04 | 2022-02-01 | 山东大学 | Point cloud registration method and system based on point cloud completion |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2015090591A (en) * | 2013-11-06 | 2015-05-11 | 株式会社パスコ | Generation device and generation method for road surface ortho-image |
| CN109493372A (en) * | 2018-10-24 | 2019-03-19 | 华侨大学 | The product point cloud data Fast global optimization method for registering of big data quantity, few feature |
| CN111860520A (en) * | 2020-07-21 | 2020-10-30 | 南京航空航天大学 | Large airplane point cloud model self-supervision semantic segmentation method based on deep learning |
| CN111915658A (en) * | 2020-09-30 | 2020-11-10 | 浙江智慧视频安防创新中心有限公司 | Registration method and system for point cloud |
| CN112017225A (en) * | 2020-08-04 | 2020-12-01 | 华东师范大学 | Depth image matching method based on point cloud registration |
| WO2021009258A1 (en) * | 2019-07-15 | 2021-01-21 | Promaton Holding B.V. | Object detection and instance segmentation of 3d point clouds based on deep learning |
| CN112365511A (en) * | 2020-11-14 | 2021-02-12 | 重庆邮电大学 | Point cloud segmentation method based on overlapped region retrieval and alignment |
-
2021
- 2021-05-13 CN CN202110521230.2A patent/CN113139991A/en active Pending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2015090591A (en) * | 2013-11-06 | 2015-05-11 | 株式会社パスコ | Generation device and generation method for road surface ortho-image |
| CN109493372A (en) * | 2018-10-24 | 2019-03-19 | 华侨大学 | The product point cloud data Fast global optimization method for registering of big data quantity, few feature |
| WO2021009258A1 (en) * | 2019-07-15 | 2021-01-21 | Promaton Holding B.V. | Object detection and instance segmentation of 3d point clouds based on deep learning |
| CN111860520A (en) * | 2020-07-21 | 2020-10-30 | 南京航空航天大学 | Large airplane point cloud model self-supervision semantic segmentation method based on deep learning |
| CN112017225A (en) * | 2020-08-04 | 2020-12-01 | 华东师范大学 | Depth image matching method based on point cloud registration |
| CN111915658A (en) * | 2020-09-30 | 2020-11-10 | 浙江智慧视频安防创新中心有限公司 | Registration method and system for point cloud |
| CN112365511A (en) * | 2020-11-14 | 2021-02-12 | 重庆邮电大学 | Point cloud segmentation method based on overlapped region retrieval and alignment |
Non-Patent Citations (1)
| Title |
|---|
| HAO XU: "OMNet: Learning Overlapping Mask for Partial-to-Partial Point Cloud Registration", 《ARXIV》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114004871A (en) * | 2022-01-04 | 2022-02-01 | 山东大学 | Point cloud registration method and system based on point cloud completion |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112288011B (en) | Image matching method based on self-attention deep neural network | |
| CN114863573B (en) | Category-level 6D attitude estimation method based on monocular RGB-D image | |
| CN112750198B (en) | Dense correspondence prediction method based on non-rigid point cloud | |
| CN113160287B (en) | Complex component point cloud splicing method and system based on feature fusion | |
| CN110490915B (en) | Point cloud registration method based on convolution-limited Boltzmann machine | |
| CN110443849B (en) | Target positioning method for double-current convolution neural network regression learning based on depth image | |
| CN112084895B (en) | Pedestrian re-identification method based on deep learning | |
| CN110310305A (en) | A kind of method for tracking target and device based on BSSD detection and Kalman filtering | |
| CN117252928B (en) | Visual image positioning system for modular intelligent assembly of electronic products | |
| Chen et al. | Full transformer framework for robust point cloud registration with deep information interaction | |
| CN113139991A (en) | 3D point cloud registration method based on overlapping region mask prediction | |
| CN114119690A (en) | Point cloud registration method based on neural network reconstruction Gaussian mixture model | |
| CN114565953A (en) | Image processing method, image processing device, electronic equipment and computer readable storage medium | |
| CN111832399A (en) | Attention mechanism fused cross-domain road navigation mark registration algorithm | |
| CN111578956A (en) | Visual SLAM positioning method based on deep learning | |
| CN115994933A (en) | Partial point cloud registration method based on consistency learning | |
| CN108469729B (en) | Human body target identification and following method based on RGB-D information | |
| CN116310128A (en) | Dynamic environment monocular multi-object SLAM method based on instance segmentation and three-dimensional reconstruction | |
| CN115829942A (en) | Electronic circuit defect detection method based on non-negative constraint sparse self-encoder | |
| CN115239776A (en) | Point cloud registration method, device, equipment and medium | |
| CN112069997B (en) | Unmanned aerial vehicle autonomous landing target extraction method and device based on DenseHR-Net | |
| CN114998630A (en) | Ground-to-air image registration method from coarse to fine | |
| De Giacomo et al. | Guided sonar-to-satellite translation | |
| CN113705731A (en) | End-to-end image template matching method based on twin network | |
| Sun et al. | Accurate deep direct geo-localization from ground imagery and phone-grade gps |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210720 |
|
| RJ01 | Rejection of invention patent application after publication |