CN109948505A - A kind of optimization method of human body three-dimensional attitude matching algorithm - Google Patents
A kind of optimization method of human body three-dimensional attitude matching algorithm Download PDFInfo
- Publication number
- CN109948505A CN109948505A CN201910191289.2A CN201910191289A CN109948505A CN 109948505 A CN109948505 A CN 109948505A CN 201910191289 A CN201910191289 A CN 201910191289A CN 109948505 A CN109948505 A CN 109948505A
- Authority
- CN
- China
- Prior art keywords
- human body
- dimensional
- posture
- matching
- attitude
- 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
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
- Processing Or Creating Images (AREA)
Abstract
The invention discloses a kind of optimization methods of human body three-dimensional attitude matching algorithm.The method comprising the steps of generation matching posture, two-dimensional projection, define loss function and pose refinement, it can be obtained by human body three-dimensional attitude matching algorithm by these steps and posture and human body three-dimensional matching posture are matched to the corresponding human body two dimension of image to be detected, then by human body three-dimensional matching posture projection, obtain human body two-dimensional projection posture, further building human body two-dimensional projection posture matches the loss function between posture with human body two dimension, by the approximate solution for solving loss function, it realizes and human body three-dimensional matching posture is optimized and revised, human body three-dimensional posture after final output optimization.This method can be combined with various human body three-dimensional attitude matching algorithms, it is versatile, and it is effective that organization of human body is combined to simplify decomposition method and iteratively faster method for solving, algorithm complexity is reduced, the precision of human body three-dimensional attitude matching algorithm is further improved.
Description
Technical field
The present invention relates to the excellent of computer image processing technology field more particularly to a kind of human body three-dimensional attitude matching algorithm
Change method.
Background technique
Human body attitude estimation is always one of computer vision and the hot issue of field of Computer Graphics, high-precision
3 D human body posture have a wide range of applications in fields such as action recognition, human-computer interaction and health cares and answered with important
With value.Currently, the two-dimension human body guise research based on convolutional neural networks have been relatively mature, the matching based on two-dimensional attitude
Class algorithm is one of 3 d pose method of mainstream.Such method is based on two-dimensional attitude testing result, in the three-dimensional pre-established
Attitude matching is carried out in attitude data library, and then obtains 3 d pose, and original image does not participate in algorithm flow directly in this method, goes
In addition to the noise jamming in original image, therefore it is more suitable for complicated actual scene.
Matching class method is to be matched based on 3 d pose database, and the 3 d pose complexity of people is higher, data
Library can not collect all 3 d poses, and the result matched is only for the posture in database, not actual scene
In situation, often not exclusively agree with real posture, still remain the lower problem of precision, be not able to satisfy and further grind
Study carefully, and does not also there is specific method to solve the problems, such as this at present.
Summary of the invention
The invention mainly solves the technical problem of providing a kind of optimization methods of human body three-dimensional attitude matching algorithm, solve
The implementation complexity of human body three-dimensional attitude matching algorithm is high in the prior art, and timeliness is not strong and exports 3 d pose precision not
High problem.
In order to solve the above technical problems, one technical scheme adopted by the invention is that providing a kind of human body three-dimensional attitude matching
The optimization method of algorithm, comprising the following steps: generate matching posture, input image to be detected, by matching class 3 D human body appearance
State algorithm obtains the corresponding human body two dimension matching posture γ of described image to be detected and human body three-dimensional matches posture y；Two-dimensional projection,
Human body three-dimensional matching posture y is projected, human body two-dimensional projection posture P (y) is obtained, P is projection camera parameter；Definition
Loss function matches posture γ with human body two dimension for human body two-dimensional projection posture P (y), defines loss letter according to Euclidean distance
Number are as follows: | | γ-P (y) | |1,L1norm；Pose refinement is realized by solving the approximate solution of the loss function to the human body
Three-dimensional matching posture y's optimizes and revises, the human body three-dimensional posture after final output optimization.
In another embodiment of optimization method of human body 3 d pose matching algorithm of the present invention, the human body three-dimensional matches appearance
State y includes K closest 3 d pose B={ b1,b2,...,bk, corresponding posture mean value is μ, further obtains human body two
Dimension projection posture P (y)=P (B α+μ), α are basic coefficient.
In another embodiment of optimization method of human body 3 d pose matching algorithm of the present invention, attitude step is matched generating
In, the corresponding organization of human body of human body attitude is decomposed into five part C={ C1,C2,C3,C4,C5, upper limb part includes C1,C2、
Lower extremities include C3,C4And torso portion C5；For each limbs chain Ci, define matrix
Jth dimension is 3x3 matrix, and codimension is set as 0, passes through EjC is indicated with yiIn j-th of joint, definition Expression endpoint is joint i1With joint i2Limbs i length, limbs length normalize standard is defined as:
In another embodiment of optimization method of human body 3 d pose matching algorithm of the present invention, the loss function is further
It may be expressed as:
In another embodiment of optimization method of human body 3 d pose matching algorithm of the present invention, in pose refinement, with most
The smallization loss function is target, and the set of the solution of artis is calculated using iterative solution method, and the set according to solution is come excellent
The each joint for changing adjustment human body three-dimensional posture, as the human body three-dimensional posture output after final optimization pass.
In another embodiment of optimization method of human body 3 d pose matching algorithm of the present invention, loss letter is minimized solving
During number, for each limbs chain Ci, include artis { (x1,y1),(x2,y2),(x3,y3), by each limbs chain Ci
It solves function and is defined as y=f (x, β), wherein β is the set of solution: β={ β1,β2,β3, i.e. joint position；To acquire optimal solution,
The residual sum of squares (RSS) S of loss function need to be made minimum, residual sum of squares (RSS) S is indicated are as follows:
In another embodiment of optimization method of human body 3 d pose matching algorithm of the present invention, in order to enable residual sum of squares (RSS)
S is minimum, and seeking partial derivative to S is 0, then has:
Wherein,Solution by iterative method is used after given initial value, wherein havingK is repeatedly
The number in generation, Δ β is iteration vector, in βkPlace can be obtained using Taylor series expansion:
Wherein,For known matrix, r is substituted the above toi=yi-f(xi, β), residual error is expressed as Δ yi=
yi-f(xi,βk),
Further neutralizing obtains:
It is converted into matrix form:
(JTJ) Δ β=JTΔg
Iterative formula is final are as follows:
JfIt is indicated for matrix of the y=f (x, β) to the first-order partial derivative with β, can show that the solution of function y=f (x, β) is close
Like pointAnd approximate solutionAnd then obtain approximate solution set β '={ β '1,β′2,β′3, foundation
Each joint position in disaggregation β ' adjustment articulated chain completes the adjustment of each articulated chain, final optimization pass entirety human body three-dimensional step by step
Posture simultaneously exports.
The beneficial effects of the present invention are: the invention discloses a kind of optimization methods of human body three-dimensional attitude matching algorithm.It should
Method includes that step has generation matching posture, two-dimensional projection, defines loss function and pose refinement, can be led to by these steps
Human body 3 d pose matching algorithm is crossed to obtain to image to be detected corresponding human body two dimension matching posture and human body three-dimensional matching appearance
State obtains human body two-dimensional projection posture, further constructs human body two-dimensional projection appearance then by human body three-dimensional matching posture projection
State matches the loss function between posture with human body two dimension, by solving the approximate solution of loss function, realizes to the human body three
Dimension matching posture is optimized and revised, the human body three-dimensional posture after final output optimization.This method can be with various human body three-dimensional appearances
State matching algorithm is combined, versatile, and effective combines that organization of human body simplifies decomposition method and iteratively faster solves
Method reduces algorithm complexity, further improves the precision of human body three-dimensional attitude matching algorithm.
Detailed description of the invention
Fig. 1 is the flow chart of one embodiment of optimization method of human body three-dimensional attitude matching algorithm according to the present invention；
Fig. 2 is the organization of human body point in another embodiment of optimization method of human body three-dimensional attitude matching algorithm according to the present invention
Solve schematic diagram；
Fig. 3 is that the process effect in another embodiment of optimization method of human body three-dimensional attitude matching algorithm according to the present invention is said
Express intention.
Specific embodiment
To facilitate the understanding of the present invention, in the following with reference to the drawings and specific embodiments, the present invention will be described in more detail.
A better embodiment of the invention is given in the attached drawing.But the invention can be realized in many different forms, and unlimited
In this specification described embodiment.On the contrary, purpose of providing these embodiments is makes to the disclosure
Understand more thorough and comprehensive.
It should be noted that unless otherwise defined, all technical and scientific terms used in this specification with belong to
The normally understood meaning of those skilled in the art of the invention is identical.Used term in the description of the invention
It is the purpose in order to describe specific embodiment, is not intended to the limitation present invention.Term "and/or" packet used in this specification
Include any and all combinations of one or more related listed items.
Fig. 1 shows the flow chart of one embodiment of optimization method of human body 3 d pose matching algorithm of the present invention.In Fig. 1
In, comprising steps of
Step S101 generates matching posture, inputs image to be detected, obtains institute by matching class 3 D human body Attitude Algorithm
State image to be detected corresponding human body two dimension matching posture γ and human body three-dimensional matching posture y；
Human body three-dimensional matching posture y is projected, obtains human body two-dimensional projection appearance by step S102, two-dimensional projection
State P (y), P are projection camera parameter；
Step S103 defines loss function, matches posture γ, root with human body two dimension for human body two-dimensional projection posture P (y)
Loss function is defined according to Euclidean distance are as follows: | | γ-P (y) | |1,L1norm；
Step S104, pose refinement are realized and are matched to the human body three-dimensional by solving the approximate solution of the loss function
Posture y's optimizes and revises, the human body three-dimensional posture after final output optimization.
Preferably, in step s101, human body three-dimensional matching posture y includes K closest 3 d pose B={ b1,
b2,...,bk, corresponding posture mean value is μ, further obtains human body two-dimensional projection posture P (y)=P (B α+μ), is based on α
Number, is calculated by posture mean μ.
Further, in step s101, the corresponding organization of human body simplification of human body attitude is decomposed into five part C=
{C1,C2,C3,C4,C5, as shown in Fig. 2, upper limb part includes C1,C2, lower extremities include C3,C4And torso portion C5；It is right
In each limbs chain Ci, define matrixJth dimension is 3x3 matrix, is indicated in the vector of joint
J-th of joint position, codimension is set as 0, indicates to exclude other joint informations, and only extracts the joint information of jth dimension,
E can be passed throughjC is indicated with yiIn j-th of joint.Definition Expression endpoint is joint i1With joint i2
The length of limbs i according to the limbs length in database, normalization standard is defined as follows in embodiments of the present invention:
Wherein, Li refers to the full-length of i-th of limbs of database.Further, for step S103, of the invention
In preferred embodiment, loss function further be may be expressed as:
Wherein γ is by obtaining the corresponding human body two of described image to be detected by matching class 3 D human body Attitude Algorithm
Dimension matching posture, L is whole limbs length.The error that we use the two-dimensional projection of 3 d pose to compare in it is as damage
Lose, specific calculation be between two-dimensional attitude corresponding joint point and two-dimensional projection's corresponding joint point of 3 d pose Euclidean away from
From summation.
Since other parameters pass through query result it is known that the only joint position one for so needing to solve in loss function
Group variable, and what we wanted optimization is also joint position, and and we are avoided by the constraint of limbs length herein
In optimization process it is possible that anti-physiology 3 d pose.
Further, in step S104, to minimize loss function as target, joint is calculated using iterative solution method
The set of the solution of point, the set according to solution is to optimize and revise each joint of 3 d pose as the human body three-dimensional after final optimization pass
Posture output.
During solving minimum loss function, for each limbs chain Ci, include artis { (x1,y1),(x2,
y2),(x3,y3), by each CiIt solves function and is defined as y=f (x, β), wherein β is the set of solution: β={ β1,β2,β3, i.e.,
Joint position.To acquire optimal solution, the residual sum of squares (RSS) S of loss function need to be made minimum, residual sum of squares (RSS) S is indicated are as follows:
Preferably, in order to enable residual sum of squares (RSS) S is minimum, seeking partial derivative to S is 0, then has:
WhereinSolution by iterative method is used after given initial value, wherein havingK is iteration
Number, Δ β be iteration vector, in βkPlace can be obtained using Taylor series expansion:
WhereinFor known matrix, r is substituted the above toi=yi-f(xi, β), residual error is expressed as Δ yi=yi-
f(xi,βk),
Further neutralizing obtains:
It is converted into matrix form:
(JTJ) Δ β=JTΔg
Iterative formula is final are as follows:
JfIt is indicated for matrix of the y=f (x, β) to the first-order partial derivative with β, i.e. Jacobian matrix, can obtain function y=f
The solution approximation point of (x, β)And approximate solutionAnd then obtain approximate solution set β '=
{β′1,β′2,β′3, according to each joint position in disaggregation β ' adjustment articulated chain, the adjustment of each articulated chain is completed step by step, finally
Optimize whole posture and exports.
It also needs to carry out a rational judgement of posture after the 3 d pose exported.Because of the target letter of optimization
The problem of number is two-dimensional attitude image, and ambiguousness is had when projecting to two-dimensional surface due to 3 d pose, that is, be parallel to
The depth of the artis of projecting direction in this direction will not influence the position of its two-dimensional projection, but position in three dimensions
It sets and is different.So we can carry out about some artis according to human body inherent structure after obtaining final 3 d pose
Beam and change, such as right and left shoulders and neck these three artis should be in same plane, and joint can not back-flexing.This
Sample can be to avoid the 3 D human body posture for not conforming to convention very much.
Fig. 3 further schematically illustrates the links of this method embodiment, can by human body three-dimensional attitude matching algorithm
To extract three-dimensional matching posture and two dimension matching posture to detection image and then be projected to obtain to three-dimensional matching posture
Two-dimensional projection's posture is established loss function between two-dimensional projection's posture and two dimension matching posture and is solved, to optimize three-dimensional
Posture output.
It can be seen that the invention discloses a kind of optimization methods of human body three-dimensional attitude matching algorithm.This method includes step
Suddenly there is generation matching posture, two-dimensional projection, define loss function and pose refinement, human body three-dimensional can be passed through by these steps
Attitude matching algorithm, which is obtained, matches posture and human body three-dimensional matching posture to the corresponding human body two dimension of image to be detected, then by people
Body three-dimensional matches posture projection, obtains human body two-dimensional projection posture, further constructs human body two-dimensional projection posture and human body two
Loss function between dimension matching posture is realized by solving the approximate solution of loss function and matches posture to the human body three-dimensional
Optimize and revise, final output optimization after human body three-dimensional posture.This method can be with various human body three-dimensional attitude matching algorithms
It is combined, it is versatile and effective that organization of human body is combined to simplify decomposition method and iteratively faster method for solving, it reduces and calculates
Method complexity further improves the precision of human body three-dimensional attitude matching algorithm.
The above description is only an embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure transformation made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant technical fields,
Similarly it is included within the scope of the present invention.
Claims (7)
1. a kind of optimization method of human body three-dimensional attitude matching algorithm, which comprises the following steps:
Matching posture is generated, image to be detected is inputted, described image to be detected is obtained by matching class 3 D human body Attitude Algorithm
Corresponding human body two dimension matching posture γ and human body three-dimensional match posture y；
Two-dimensional projection projects human body three-dimensional matching posture y, obtains human body two-dimensional projection posture P (y), and P is projection
Camera parameter；
Loss function is defined, posture γ is matched with human body two dimension for human body two-dimensional projection posture P (y), it is fixed according to Euclidean distance
Adopted loss function are as follows: | | γ-P (y) | |1,L1norm；
Pose refinement realizes the optimization tune to human body three-dimensional matching posture y by solving the approximate solution of the loss function
Whole, after final output optimization human body three-dimensional posture.
2. the optimization method of human body three-dimensional attitude matching algorithm according to claim 1, which is characterized in that the human body three
Dimension matching posture y includes K closest 3 d pose B={ b1,b2,...,bk, corresponding posture mean value is μ, further
It is basic coefficient to human body two-dimensional projection posture P (y)=P (B α+μ), α.
3. the optimization method of human body three-dimensional attitude matching algorithm according to claim 2, which is characterized in that matched generating
In attitude step, the corresponding organization of human body of human body attitude is decomposed into five part C={ C1,C2,C3,C4,C5, upper limb part packet
Include C1,C2, lower extremities include C3,C4And torso portion C5；For each limbs chain Ci, define matrixJth dimension is 3x3 matrix, and codimension is set as 0, passes through EjC is indicated with yiIn j-th pass
Section, definition Expression endpoint is joint i1With joint i2Limbs i length, limbs length normalization
Standard is defined as:
4. the optimization method of human body three-dimensional attitude matching algorithm according to claim 3, which is characterized in that the loss letter
Number further may be expressed as:
5. the optimization method of human body three-dimensional attitude matching algorithm according to claim 4, which is characterized in that in pose refinement
In, to minimize the loss function as target, the set of the solution of artis, the collection according to solution are calculated using iterative solution method
It closes to optimize and revise each joint of human body three-dimensional posture, as the human body three-dimensional posture output after final optimization pass.
6. the optimization method of human body three-dimensional attitude matching algorithm according to claim 5, which is characterized in that minimum solving
During changing loss function, for each limbs chain Ci, include artis { (x1,y1),(x2,y2),(x3,y3), it will be each
A limbs chain CiIt solves function and is defined as y=f (x, β), wherein β is the set of solution: β={ β1,β2,β3, i.e. joint position；For
Optimal solution is acquired, the residual sum of squares (RSS) S of loss function need to be made minimum, residual sum of squares (RSS) S is indicated are as follows:
7. the optimization method of human body three-dimensional attitude matching algorithm according to claim 6, which is characterized in that in order to enable residual
Poor quadratic sum S is minimum, and seeking partial derivative to S is 0, then has:
Wherein,Solution by iterative method is used after given initial value, wherein havingK is iteration
Number, Δ β is iteration vector, in βkPlace can be obtained using Taylor series expansion:
Wherein,For known matrix, r is substituted the above toi=yi-f(xi, β), residual error is expressed as Δ yi=yi-f
(xi,βk),
Further neutralizing obtains:
It is converted into matrix form:
(JTJ) Δ β=JTΔg
Iterative formula is final are as follows:
JfIt is indicated for matrix of the y=f (x, β) to the first-order partial derivative with β, can obtain the solution approximation point of function y=f (x, β)
And approximate solutionAnd then obtain approximate solution set β '={ β '1,β'2,β'3, according to disaggregation β '
Each joint position in articulated chain is adjusted, completes the adjustment of each articulated chain step by step, final optimization pass entirety human body 3 d pose is simultaneously
Output.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910191289.2A CN109948505B (en) | 2019-03-14 | 2019-03-14 | Optimization method of human body three-dimensional posture matching algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910191289.2A CN109948505B (en) | 2019-03-14 | 2019-03-14 | Optimization method of human body three-dimensional posture matching algorithm |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109948505A true CN109948505A (en) | 2019-06-28 |
| CN109948505B CN109948505B (en) | 2021-03-30 |
Family
ID=67009829
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910191289.2A Active CN109948505B (en) | 2019-03-14 | 2019-03-14 | Optimization method of human body three-dimensional posture matching algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109948505B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111291656A (en) * | 2020-01-21 | 2020-06-16 | 杭州微洱网络科技有限公司 | Human body posture matching method for 2d image measurement in E-commerce image |
| CN111652113A (en) * | 2020-05-29 | 2020-09-11 | 北京百度网讯科技有限公司 | Obstacle detection method, apparatus, device, and storage medium |
| CN113449565A (en) * | 2020-03-27 | 2021-09-28 | 海信集团有限公司 | Three-dimensional attitude estimation method, intelligent device and storage medium |
| CN114332939A (en) * | 2021-12-30 | 2022-04-12 | 浙江核新同花顺网络信息股份有限公司 | Pose sequence generation method and system |
| WO2022115991A1 (en) * | 2020-12-01 | 2022-06-09 | Intel Corporation | Incremental 2d-to-3d pose lifting for fast and accurate human pose estimation |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103155003A (en) * | 2010-10-08 | 2013-06-12 | 松下电器产业株式会社 | Posture estimation device and posture estimation method |
| CN107886089A (en) * | 2017-12-11 | 2018-04-06 | 深圳市唯特视科技有限公司 | A kind of method of the 3 D human body Attitude estimation returned based on skeleton drawing |
| CN108022278A (en) * | 2017-12-29 | 2018-05-11 | 清华大学 | Character animation method for drafting and system based on motion tracking in video |
| CN108629801A (en) * | 2018-05-14 | 2018-10-09 | 华南理工大学 | A kind of three-dimensional (3 D) manikin posture of video sequence and Shape Reconstruction method |
| CN108648203A (en) * | 2018-04-24 | 2018-10-12 | 上海工程技术大学 | A method of the human body three-dimensional Attitude estimation based on monocular cam |
-
2019
- 2019-03-14 CN CN201910191289.2A patent/CN109948505B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103155003A (en) * | 2010-10-08 | 2013-06-12 | 松下电器产业株式会社 | Posture estimation device and posture estimation method |
| US20130230211A1 (en) * | 2010-10-08 | 2013-09-05 | Panasonic Corporation | Posture estimation device and posture estimation method |
| CN107886089A (en) * | 2017-12-11 | 2018-04-06 | 深圳市唯特视科技有限公司 | A kind of method of the 3 D human body Attitude estimation returned based on skeleton drawing |
| CN108022278A (en) * | 2017-12-29 | 2018-05-11 | 清华大学 | Character animation method for drafting and system based on motion tracking in video |
| CN108648203A (en) * | 2018-04-24 | 2018-10-12 | 上海工程技术大学 | A method of the human body three-dimensional Attitude estimation based on monocular cam |
| CN108629801A (en) * | 2018-05-14 | 2018-10-09 | 华南理工大学 | A kind of three-dimensional (3 D) manikin posture of video sequence and Shape Reconstruction method |
Non-Patent Citations (1)
| Title |
|---|
| J.WANG ET AL.: "《 Fall Detection Using a Multistage Deep Convolutional Network Architecture》", 《LECTURE NOTES IN ELECTRICAL ENGINEERING (LNEE 536)》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111291656A (en) * | 2020-01-21 | 2020-06-16 | 杭州微洱网络科技有限公司 | Human body posture matching method for 2d image measurement in E-commerce image |
| CN111291656B (en) * | 2020-01-21 | 2023-06-02 | 杭州微洱网络科技有限公司 | Human body trunk posture matching method in measurement 2d image |
| CN113449565A (en) * | 2020-03-27 | 2021-09-28 | 海信集团有限公司 | Three-dimensional attitude estimation method, intelligent device and storage medium |
| CN111652113A (en) * | 2020-05-29 | 2020-09-11 | 北京百度网讯科技有限公司 | Obstacle detection method, apparatus, device, and storage medium |
| US11688177B2 (en) | 2020-05-29 | 2023-06-27 | Apollo Intelligent Connectivity (Beijing) Technology Co., Ltd. | Obstacle detection method and device, apparatus, and storage medium |
| WO2022115991A1 (en) * | 2020-12-01 | 2022-06-09 | Intel Corporation | Incremental 2d-to-3d pose lifting for fast and accurate human pose estimation |
| CN114332939A (en) * | 2021-12-30 | 2022-04-12 | 浙江核新同花顺网络信息股份有限公司 | Pose sequence generation method and system |
| CN114332939B (en) * | 2021-12-30 | 2024-02-06 | 浙江核新同花顺网络信息股份有限公司 | Pose sequence generation method and system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109948505B (en) | 2021-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948505A (en) | A kind of optimization method of human body three-dimensional attitude matching algorithm | |
| Deprelle et al. | Learning elementary structures for 3d shape generation and matching | |
| Sarode et al. | Pcrnet: Point cloud registration network using pointnet encoding | |
| Pishchulin et al. | Building statistical shape spaces for 3d human modeling | |
| CN107102727B (en) | Dynamic gesture learning and recognition method based on ELM neural network | |
| CN111627101B (en) | Three-dimensional human reconstruction method based on graph convolution | |
| Lei et al. | Cadex: Learning canonical deformation coordinate space for dynamic surface representation via neural homeomorphism | |
| CN108629294A (en) | Human body based on deformation pattern and face net template approximating method | |
| Moxon et al. | Improved Cauchy-characteristic evolution system for high-precision numerical relativity waveforms | |
| Luo et al. | Gait recognition using GEI and AFDEI | |
| Huang et al. | Parametric design for human body modeling by wireframe-assisted deep learning | |
| JPWO2019030794A1 (en) | Information processing device, model data creation program, model data creation method | |
| JP2023524252A (en) | Generative nonlinear human shape model | |
| CN106778708A (en) | A kind of expression shape change recognition methods of the active appearance models based on tensor | |
| CN113538608A (en) | Controllable character image generation method based on generation countermeasure network | |
| Yuan et al. | EGST: Enhanced Geometric Structure Transformer for Point Cloud Registration | |
| Laga et al. | 4d atlas: Statistical analysis of the spatiotemporal variability in longitudinal 3D shape data | |
| CN108876837A (en) | One kind being based on L1/2The 3 D human body attitude reconstruction method of regularization | |
| Taheri et al. | Joint albedo estimation and pose tracking from video | |
| Lin et al. | Neural-network-based adaptive hybrid-reflectance model for 3-D surface reconstruction | |
| Xu et al. | ScaleLK: registration of point clouds with different scales using deep learning methods | |
| Xia et al. | Keyframe-editable real-time motion synthesis | |
| Zhao et al. | Posenormnet: Identity-preserved posture normalization of 3d body scans in arbitrary postures | |
| Wang et al. | Non-rigid object alignment with a mismatch template based on exhaustive local search | |
| CN111415377A (en) | Partial quality optimal transmission theory-based incomplete point cloud registration method |
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 |