CN113222943B - Image deformation estimation method based on mixed grid transformation model - Google Patents
Image deformation estimation method based on mixed grid transformation model Download PDFInfo
- Publication number
- CN113222943B CN113222943B CN202110539834.XA CN202110539834A CN113222943B CN 113222943 B CN113222943 B CN 113222943B CN 202110539834 A CN202110539834 A CN 202110539834A CN 113222943 B CN113222943 B CN 113222943B
- Authority
- CN
- China
- Prior art keywords
- grid
- matrix
- characteristic point
- transformation model
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- 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
- G06T7/337—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods involving reference images or patches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- 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
- G06T2207/10004—Still image; Photographic image
Abstract
An image deformation estimation method based on a hybrid grid transformation model belongs to the field of computer vision and image processing. The method solves the problem that the inter-image transformation model obtained by the existing estimation method cannot effectively solve the discontinuous deformation. Firstly, extracting and matching feature points of an image; then, carrying out primary classification and screening on the characteristic point pairs; and finally, estimating a hybrid grid transformation model. The method can accurately estimate the mixed grid transformation model and the component number of the mixed grid transformation model, and the estimated mixed grid transformation model can effectively solve the problem of discontinuous deformation. The invention can be used for estimating the image deformation.
Description
Technical Field
The invention belongs to the field of computer vision and image processing, and particularly relates to an image deformation estimation method based on a hybrid grid transformation model.
Background
Image registration is a common problem in the field of computer vision and image processing, and can be applied to occasions such as image splicing, target positioning, three-dimensional reconstruction and the like. When an image of a real scene is shot, due to the fact that the distances between the positions of different objects and a camera are different, object displacement discontinuity exists in the images shot from different viewpoints, image registration operation needs to be conducted on the images shot from the different viewpoints, but the problem of discontinuous deformation cannot be effectively solved through an inter-image transformation model obtained through an existing estimation method.
Disclosure of Invention
The invention aims to solve the problem that a transformation model between images obtained by the existing estimation method cannot effectively solve the discontinuous deformation, and provides an image deformation estimation method based on a mixed grid transformation model.
The technical scheme adopted by the invention for solving the technical problems is as follows: an image deformation estimation method based on a hybrid mesh transformation model specifically comprises the following steps:
step one, respectively extracting reference images I of the same scene shot under different visual angles0And a target image I1And for reference image I0SIFT feature point and target image I1The SIFT feature points are matched to obtain an initial feature point pair set S0,Wherein the content of the first and second substances,as the pair of initial characteristic points,for reference picture I0The coordinates of the feature point in (2),is a target image I1Characteristic point coordinates of (1), N0Is a set S0The number of the middle initial characteristic point pairs;
step two, classifying and screening the initial characteristic point pairs obtained in the step one, wherein the set of the characteristic point pairs obtained through screening is S,(xi,yi) For pairs of characteristic points, x, obtained by screeningiFor reference picture I0Coordinates of the feature points in (1), yiIs a target image I1The coordinate of the characteristic point in the set S, N is the number of the characteristic point pairs in the set S, and the characteristic point coordinates in the set S are obtainedThe total number of categories of characteristic point pairs included in the set S is represented as K, and the category of the ith characteristic point pair is represented as c for each category of characteristic point pairi,ci=1,...,K;
Step three, for the reference image I0Carrying out grid division, wherein the width of each grid is w, and the height of each grid is h; taking a mixed grid transformation model consisting of K grid transformation models as a reference image I0And a target image I1Inter-geometric transformation models, each mesh transformation model constituting a component of the hybrid mesh transformation model;
wherein, the grid deformation parameter to be solved of each component is wk,k=1,2,…,K,wkIs the initial covariance matrix of Is a matrix of0The inverse matrix of (d);
step four, respectively calculating each characteristic point xiCorresponding intermediate variable matrix phii;
Initializing a responsibility matrix R, wherein the size of R is NxK, and the ith row and kth column elements in the responsibility matrix R are Ri,kIf k is equal to ciThen r isi,k1, otherwise ri,k0; initializing the iteration number n to be 0;
step six, for each component, utilizing phiiAnd R calculates wkMean value m of the posterior probability distribution ofkCovariance matrixAnd a parameter alphak;
step seven, utilizing mk、And alphakUpdating element values R in the responsibility matrix Ri,kThe updated element value is r'i,k;
And step eight, updating the iteration number n to n +1 according to the element value r 'updated in the step seven'i,kComputing
Step nine, repeating the process from the step six to the step eight until n is more than 1 andstopping iteration and obtaining m in the last iterationkAs wkAn estimate of (d).
The invention has the beneficial effects that: the invention provides an image deformation estimation method based on a hybrid grid transformation model, which comprises the steps of firstly extracting and matching characteristic points of an image; then, carrying out primary classification and screening on the characteristic point pairs; and finally, estimating a hybrid grid transformation model. The method can accurately estimate the mixed grid transformation model and the component number of the mixed grid transformation model, and the estimated mixed grid transformation model can effectively solve the problem of discontinuous deformation.
Drawings
FIG. 1 is a flow chart of an image deformation estimation method based on a hybrid mesh transformation model according to the present invention;
FIG. 2a) is the same scene graph one taken at a different viewpoint;
FIG. 2b) is the same scene graph two taken at a different viewpoint;
the characteristic points and grid lines of each component of the hybrid grid transformation model are represented in the figure.
Detailed Description
First embodiment this embodiment will be described with reference to fig. 1. The method for estimating image deformation based on a hybrid mesh transformation model according to the present embodiment specifically includes the following steps:
step one, respectively extracting reference images I of the same scene shot under different visual angles0And a target image I1And for reference image I0SIFT feature point and target image I1The SIFT feature points are matched to obtain an initial feature point pair set S0,Wherein the content of the first and second substances,is the initial characteristic point pair, and the characteristic point pair,for reference picture I0The coordinates of the feature points in (1),is a target image I1Characteristic point coordinates of (1), N0Is a set S0The number of the middle initial characteristic point pairs;
SIFT is a commonly used feature point extraction and description algorithm, has the invariance of illumination, rotation and deformation to a certain degree, and is a commonly used method because the extracted feature points are stable and have high quality;
step two, classifying and screening the initial characteristic point pairs obtained in the step one, wherein the set of the characteristic point pairs obtained through screening isS,(xi,yi) For pairs of characteristic points, x, obtained by screeningiFor reference picture I0Coordinates of the feature points in (1), yiIs a target image I1N is the number of characteristic point pairs in the set S, and the category of each characteristic point pair in the set S is obtained, the total number of categories of characteristic point pairs included in the set S is represented as K, and the category of the ith characteristic point pair is represented as ci,ci=1,...,K;
Step three, as shown in fig. 2a) and 2 b); for reference image I0Carrying out grid division, wherein the width of each grid is w, and the height of each grid is h; taking a mixed grid transformation model consisting of K grid transformation models as a reference image I0And a target image I1Inter-geometric transformation models, each mesh transformation model constituting a component of the hybrid mesh transformation model;
wherein, the grid deformation parameter to be solved of each component is wk,k=1,2,…,K,wkIs the initial covariance matrix ofIs a matrix of0The inverse matrix of (d);
step four, respectively calculating each characteristic point xiCorresponding intermediate variable matrix phiiPhi matrixiThe size of (a) is 2M × 2;
initializing a responsibility matrix R, wherein the size of R is NxK, and the ith row and kth column elements in the responsibility matrix R are Ri,kIf k is equal to ciThen r isi,k1, otherwise ri,k0; initializing the iteration number n to be 0;
step six, for each component, utilizing phiiAnd R calculates wkMean value m of the posterior probability distribution ofkCovariance matrixAnd a parameter alphak;
step seven, utilizing mk、And alphakUpdating element values R in the responsibility matrix Ri,kThe updated element value is r'i,k;
And step eight, updating the iteration number n to n +1 according to the element value r 'updated in the step seven'i,kComputing
Step nine, repeating the process from the step six to the step eight until n is more than 1 andstopping iteration and obtaining m in the last iterationkAs wkAn estimate of (d).
I.e. when n > 1 and all components are satisfied simultaneouslyThen m obtained in the last iteration is addedkAs wkIs estimated value of. In the iteration process, the element value R in the updated responsibility matrix R obtained in the iteration is usedi,kFor the next iteration process.
Estimating the model by the expectation maximization iteration step has the advantages of avoiding overfitting and automatically determining the number of model components.
The second embodiment is as follows: the difference between this embodiment and the first embodiment is that, in the first step, the reference image I is searched based on the FLANN fast nearest neighbor search library0SIFT feature point and target image I1The SIFT feature points are matched.
The FLANN fast nearest neighbor search library can realize fast nearest neighbor matching of high-dimensional feature points.
The third concrete implementation mode: the difference between this embodiment and the second embodiment is that the method for classifying and screening the initial feature point pairs obtained in the first step is a RANSAC algorithm.
RANSAC is a parameterized model estimation algorithm and is characterized in that the method is robust to abnormal points and can obtain a correct model of effective sample data.
The fourth concrete implementation mode is as follows: the third difference between this embodiment and the specific embodiment is that the classification and screening of the initial feature point pairs obtained in step one includes the following specific processes:
step two, setting the current residual characteristic point pair set as S', and initializing the set as S0(ii) a Representing the characteristic point pair set obtained through screening as S, and initializing the set S into an empty set; initializing the serial number of the current category as c' ═ 1;
secondly, extracting interior points from the set S' by adopting RANSAC algorithm to obtain an interior point set Sc′Setting an inner point distance threshold value as Q and iteration times as L;
step two and three, if Sc′If the number of the characteristic point pairs is more than or equal to 10, updating S to S + Sc′And will aggregate Sc′Setting the class number of the middle characteristic point pair to be c ' and updating S ' ═ S ' -Sc′,c′=c′+1;
The second step four,Repeatedly executing the second step and the third step until S appearsc′If the number of the characteristic point pairs is less than 10, the process is ended.
The fifth concrete implementation mode: the fourth difference between this embodiment and the fourth embodiment is that the model selected by the RANSAC algorithm is a homography matrix transformation model.
The sixth specific implementation mode: the fourth difference between this embodiment and the specific embodiment is that the value of the interior point distance threshold Q is 3, and the value of the iteration number L is 500.
The seventh embodiment: the sixth embodiment is different from the sixth embodiment in that the matrix Λ0The expression of (a) is:
Λ0=cUTU
wherein c is a constant, and the value is 5 in the invention; u is a sparse matrix of 2 Mx 2M size, M is for reference image I0The number of grid points obtained by grid division and the element U of the mth row and mth column in the sparse matrix Um,m1, mth row of UElements of a column The number of the grid points connected to the mth grid point is the serial number of the grid point connected to the mth grid point (according to the actual situation, the number of the grid points connected to the mth grid point may be 2,3, or 4), and the remaining elements in U are all 0; the superscript T stands for transpose.
The specific implementation mode is eight: the difference between this embodiment and the seventh embodiment is that the specific process of step four is as follows:
characteristic point xiThe sequence numbers of 4 grid points of the grid areThe bilinear interpolation weights of 1,2,3,4 and 4 grid points are respectivelyThen the intermediate variable matrix phiiTo (1) aThe row column 1 element has a value ofFirst, theRow column 2 element has a value ofThe values of the remaining elements are all 0.
In the present invention, the numbers of the grid points may be arranged in any manner, but it is only required that the grid point numbers in this embodiment and the grid point numbers in the seventh embodiment are obtained in the same arrangement manner.
The specific implementation method nine: the present embodiment is different from the first embodiment in that the value of the constant β is 0.05.
The detailed implementation mode is ten: in this embodiment, the constant α is different from the first embodiment0Is 0.01.
The concrete implementation mode eleven: the difference between this embodiment and the eighth embodiment is that the specific process of the seventh step is as follows:
where ρ isi,kIs an intermediate variable;
where ψ (-) is a double gamma function and tr (-) is the trace of the matrix.
The above-described calculation examples of the present invention are merely to explain the calculation model and the calculation flow of the present invention in detail, and are not intended to limit the embodiments of the present invention. It will be apparent to those skilled in the art that other variations and modifications of the present invention can be made based on the above description, and it is not intended to be exhaustive or to limit the invention to the precise form disclosed, and all such modifications and variations are possible and contemplated as falling within the scope of the invention.
Claims (9)
1. An image deformation estimation method based on a hybrid mesh transformation model is characterized by specifically comprising the following steps:
step one, respectively extracting reference images I of the same scene shot under different visual angles0And a target image I1And for reference image I0SIFT feature point and target image I1The SIFT feature points are matched to obtain an initial feature point pair set S0,Wherein the content of the first and second substances,as the pair of initial characteristic points,for reference picture I0The coordinates of the feature points in (1),is a target image I1Characteristic point coordinates of (1), N0Is a set S0The number of the middle initial characteristic point pairs;
step two, classifying and screening the initial characteristic point pairs obtained in the step one, wherein the set of the characteristic point pairs obtained through screening is S,(xi,yi) For pairs of characteristic points, x, obtained by screeningiFor reference picture I0Coordinates of the feature points in (1), yiIs a target image I1N is the number of characteristic point pairs in the set S, and the category of each characteristic point pair in the set S is obtained, the total number of categories of characteristic point pairs included in the set S is represented as K, and the category of the ith characteristic point pair is represented as ci,ci=1,...,K;
Step three, for the reference image I0Carrying out grid division, wherein the width of each grid is w, and the height of each grid is h; taking a mixed grid transformation model consisting of K grid transformation models as a reference image I0And a target image I1Inter-geometric transformation models, each mesh transformation model constituting a component of the hybrid mesh transformation model;
wherein, the grid deformation parameter to be solved of each component is wk,k=1,2,…,K,wkIs the initial covariance matrix of Is a matrix of0The inverse matrix of (d);
step four, respectively calculating each characteristic point xiCorresponding intermediate variable matrix phii;
The specific process of the step four is as follows:
characteristic point xiThe sequence numbers of the 4 grid points of the grid areThe bilinear interpolation weights of the 4 grid points are respectivelyThen the intermediate variable matrix phiiTo (1) aThe row column 1 element has a value ofFirst, theRow column 2 element has a value ofThe values of the other elements are all 0;
initializing a responsibility matrix R, wherein the size of R is NxK, and the ith row and kth column elements in the responsibility matrix R are Ri,kIf k is equal to ciThen r isi,k1, otherwise ri,k0; initializing the iteration number n to be 0;
step six, for each component, utilizing phiiAnd R calculates wkMean value m of the posterior probability distribution ofkCovariance matrixAnd a parameter alphak;
step seven, utilizing mk、And alphakUpdating element values R in the responsibility matrix Ri,kWhere the updated element value is denoted by r'i,k;
The concrete process of the seventh step is as follows:
where ρ isi,kIs an intermediate variable;
wherein psi (-) is a double gamma function, and tr (-) is the trace of the matrix;
and step eight, updating the iteration number n to n +1 according to the element value r 'updated in the step seven'i,kComputing
2. The method according to claim 1, wherein in the first step, the reference image I is estimated based on a FLANN fast nearest neighbor search library0SIFT feature point and target image I1The SIFT feature points are matched.
3. The method for estimating image deformation based on hybrid mesh transform model as claimed in claim 2, wherein the initial feature point pairs obtained in the first step are classified and screened by using RANSAC algorithm.
4. The method according to claim 3, wherein the classifying and screening of the initial feature point pairs obtained in the first step comprises:
step two, setting the current residual feature point pair set as S', and initializing the feature point pair set as S0(ii) a Representing the characteristic point pair set obtained through screening as S, and initializing the set S into an empty set; initializing the serial number of the current category as c' ═ 1;
secondly, extracting interior points from the set S' by adopting RANSAC algorithm to obtain an interior point set Sc′Setting an inner point distance threshold value as Q and iteration times as L;
step two and step three, if Sc′If the number of the characteristic point pairs is more than or equal to 10, updating S to S + Sc′And will aggregate Sc′Setting the class number of the middle characteristic point pair to be c ' and updating S ' ═ S ' -Sc′,c′=c′+1;
Step two and step four, repeatedly executing step two and step three until S appearsc′If the number of the characteristic point pairs is less than 10, the process is ended.
5. The method for image deformation estimation based on hybrid grid transformation model as claimed in claim 4, wherein the RANSAC algorithm uses a homography matrix transformation model to classify and screen the initial feature point pairs.
6. The image deformation estimation method based on the hybrid mesh transformation model according to claim 4, wherein the value of the interior point distance threshold Q is 3, and the value of the iteration number L is 500.
7. The image deformation estimation method based on the hybrid mesh transformation model as claimed in claim 6, wherein the matrix Λ is0The expression of (a) is:
Λ0=cUTU
wherein c is a constant; u is a sparse matrix of 2 Mx 2M size, M is for reference image I0The number of grid points obtained by grid division and the element U of the mth row and mth column in the sparse matrix Um,m1, mth row of UElements of a column The number of the grid point connected with the mth grid point is the serial number of the grid point, and the rest elements in the U are 0; the superscript T stands for transpose.
8. The image deformation estimation method based on the hybrid mesh transform model according to claim 1, wherein the value of the constant β is 0.05.
9. The method of claim 1, wherein the constant α is the same as the constant α0Is 0.01.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110539834.XA CN113222943B (en) | 2021-05-18 | 2021-05-18 | Image deformation estimation method based on mixed grid transformation model |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110539834.XA CN113222943B (en) | 2021-05-18 | 2021-05-18 | Image deformation estimation method based on mixed grid transformation model |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113222943A CN113222943A (en) | 2021-08-06 |
CN113222943B true CN113222943B (en) | 2022-05-03 |
Family
ID=77092573
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110539834.XA Active CN113222943B (en) | 2021-05-18 | 2021-05-18 | Image deformation estimation method based on mixed grid transformation model |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113222943B (en) |
Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5651075A (en) * | 1993-12-01 | 1997-07-22 | Hughes Missile Systems Company | Automated license plate locator and reader including perspective distortion correction |
CN101221658A (en) * | 2007-12-20 | 2008-07-16 | 四川川大智胜软件股份有限公司 | Cylinder frame buffer texture re-labeling geometric correction method based on software |
CN101572828A (en) * | 2009-05-20 | 2009-11-04 | 长春理工大学 | Method for correcting distortion in real time based on GPU camera and video camera |
CN102194212A (en) * | 2010-03-08 | 2011-09-21 | 佳能株式会社 | Image processing method, device and system |
CN103729849A (en) * | 2013-12-31 | 2014-04-16 | 南京航空航天大学 | Method for calculating digital image morphing initial value |
CN103729841A (en) * | 2013-12-18 | 2014-04-16 | 同济大学 | Camera distortion correcting method based on square target model and perspective projection |
CN104574363A (en) * | 2014-12-12 | 2015-04-29 | 南京邮电大学 | Full reference image quality assessment method in consideration of gradient direction difference |
CN104994283A (en) * | 2015-06-30 | 2015-10-21 | 广东欧珀移动通信有限公司 | Correction method for local distortion and mobile terminal |
CN108780586A (en) * | 2016-03-09 | 2018-11-09 | 株式会社理光 | Image processing method, display equipment and inspection system |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
AU2016266019A1 (en) * | 2016-11-30 | 2018-06-14 | Canon Kabushiki Kaisha | Image registration method |
-
2021
- 2021-05-18 CN CN202110539834.XA patent/CN113222943B/en active Active
Patent Citations (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5651075A (en) * | 1993-12-01 | 1997-07-22 | Hughes Missile Systems Company | Automated license plate locator and reader including perspective distortion correction |
CN101221658A (en) * | 2007-12-20 | 2008-07-16 | 四川川大智胜软件股份有限公司 | Cylinder frame buffer texture re-labeling geometric correction method based on software |
CN101572828A (en) * | 2009-05-20 | 2009-11-04 | 长春理工大学 | Method for correcting distortion in real time based on GPU camera and video camera |
CN102194212A (en) * | 2010-03-08 | 2011-09-21 | 佳能株式会社 | Image processing method, device and system |
CN103729841A (en) * | 2013-12-18 | 2014-04-16 | 同济大学 | Camera distortion correcting method based on square target model and perspective projection |
CN103729849A (en) * | 2013-12-31 | 2014-04-16 | 南京航空航天大学 | Method for calculating digital image morphing initial value |
CN104574363A (en) * | 2014-12-12 | 2015-04-29 | 南京邮电大学 | Full reference image quality assessment method in consideration of gradient direction difference |
CN104994283A (en) * | 2015-06-30 | 2015-10-21 | 广东欧珀移动通信有限公司 | Correction method for local distortion and mobile terminal |
CN108780586A (en) * | 2016-03-09 | 2018-11-09 | 株式会社理光 | Image processing method, display equipment and inspection system |
Non-Patent Citations (3)
Title |
---|
Distorted Building Image Matching with Automatic Viewpoint Rectification and Fusion;Linwei Yue 等;《MDPI》;20191127;全文 * |
变形图驱动的人体与脸部网格模板拟合系统;黎子聪 等;《计算机辅助设计与图形学学报》;20190731;全文 * |
基于图像块分配的鲁棒模板匹配方法;张智浩 等;《2020中国自动化大会(CA2020)论文集》;20210131;全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN113222943A (en) | 2021-08-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108038435B (en) | Feature extraction and target tracking method based on convolutional neural network | |
JP6395555B2 (en) | How to process video captured from a scene | |
CN112328715B (en) | Visual positioning method, training method of related model, related device and equipment | |
CN107506792B (en) | Semi-supervised salient object detection method | |
CN111768415A (en) | Image instance segmentation method without quantization pooling | |
CN110490894B (en) | Video foreground and background separation method based on improved low-rank sparse decomposition | |
Temel et al. | CSV: Image quality assessment based on color, structure, and visual system | |
CN113177592B (en) | Image segmentation method and device, computer equipment and storage medium | |
CN112215079B (en) | Global multistage target tracking method | |
CN112084952B (en) | Video point location tracking method based on self-supervision training | |
CN111127353B (en) | High-dynamic image ghost-removing method based on block registration and matching | |
CN107392211B (en) | Salient target detection method based on visual sparse cognition | |
CN112784747B (en) | Multi-scale eigen decomposition method for hyperspectral remote sensing image | |
CN104680181B (en) | SAR image superpixel segmentation method based on likelihood ratio feature | |
CN117036756B (en) | Remote sensing image matching method and system based on variation automatic encoder | |
CN111161348A (en) | Monocular camera-based object pose estimation method, device and equipment | |
CN110738695B (en) | Image feature point mismatching and removing method based on local transformation model | |
CN113222943B (en) | Image deformation estimation method based on mixed grid transformation model | |
CN110136164B (en) | Method for removing dynamic background based on online transmission transformation and low-rank sparse matrix decomposition | |
CN111814884A (en) | Target detection network model upgrading method based on deformable convolution | |
Zhou et al. | Stn-homography: estimate homography parameters directly | |
CN108765384B (en) | Significance detection method for joint manifold sequencing and improved convex hull | |
CN110570450A (en) | Target tracking method based on cascade context-aware framework | |
WO2022239216A1 (en) | Training device, training method, image processing device, and image processing method | |
CN113920391A (en) | Target counting method based on generated scale self-adaptive true value graph |
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 |