CN111862313A - Energy optimization-based panda three-dimensional model reconstruction method - Google Patents

Energy optimization-based panda three-dimensional model reconstruction method Download PDF

Info

Publication number
CN111862313A
CN111862313A CN202010715308.XA CN202010715308A CN111862313A CN 111862313 A CN111862313 A CN 111862313A CN 202010715308 A CN202010715308 A CN 202010715308A CN 111862313 A CN111862313 A CN 111862313A
Authority
CN
China
Prior art keywords
panda
dimensional
dimensional model
representing
projection
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.)
Withdrawn
Application number
CN202010715308.XA
Other languages
Chinese (zh)
Inventor
胡绍湘
陈鹏
罗敏
侯蓉
蒋岚
廖志武
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
University of Electronic Science and Technology of China
Original Assignee
University of Electronic Science and Technology of China
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by University of Electronic Science and Technology of China filed Critical University of Electronic Science and Technology of China
Priority to CN202010715308.XA priority Critical patent/CN111862313A/en
Publication of CN111862313A publication Critical patent/CN111862313A/en
Withdrawn legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects

Landscapes

  • Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Graphics (AREA)
  • Geometry (AREA)
  • Software Systems (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Length Measuring Devices By Optical Means (AREA)
  • Image Analysis (AREA)

Abstract

The invention provides a panda three-dimensional model reconstruction method based on energy optimization, and belongs to the field of three-dimensional reconstruction. Inputting a two-dimensional image of a panda; acquiring two-dimensional key points of the pandas and image contour segmentation, and setting the corresponding relation between the key points and the vertex of the three-dimensional model; utilizing a prior information panda SMAL model to control the deformation of the panda three-dimensional model by updating shape and posture parameters; carrying out perspective projection on the deformation model to obtain projection key points and a projection area, and constructing a total energy optimization function; minimizing an energy optimization function, and iteratively updating the shape parameters and the attitude parameters; and combining the parameters with the SMAL model to obtain the final three-dimensional reconstruction result of the pandas. The method can directly reconstruct the three-dimensional model of the panda from the two-dimensional image by combining the prior information of the panda SMAL model in an energy optimization mode, and fills the blank of the field of the three-dimensional reconstruction of the panda.

Description

Energy optimization-based panda three-dimensional model reconstruction method
Technical Field
The invention belongs to the field of three-dimensional reconstruction, and particularly relates to a panda three-dimensional model reconstruction method based on energy optimization.
Background
Three-dimensional reconstruction is a popular research direction in the field of computer vision, and is widely applied to the fields of automatic driving, digital archaeology, medical three-dimensional imaging and the like. The pandas are national treasures in China, the first-class protection of animals in China, and panda research bases are set up in many places in China for research and protection of pandas, but the three-dimensional reconstruction of pandas is a relatively blank field at present. Through the deep research on the giant panda three-dimensional reconstruction technology, the spatial morphological structure of the giant panda can be truly and completely reflected, the giant panda model is spread and displayed in the most intuitive form, the body ruler measurement mode of the giant panda can be changed from contact to non-contact, the cost is reduced, the resource is saved, the efficiency is improved, the measurement of the height, the length, the chest circumference and the like of the giant panda is completed, the development conditions of the giant panda, such as high, short and thin bodies, weight increase and decrease, body length increase and decrease and the like, are more clearly known, the living environment and the health condition of the giant panda are further analyzed, the species of the giant panda are better protected, and the protection level of the giant panda is improved. Meanwhile, through rapid three-dimensional reconstruction, three-dimensional display and the like, people can know and know the pandas more intuitively and comprehensively in a flexible and various form, and the animal protection consciousness of the people is further enhanced through an interesting interaction mode.
Currently, three-dimensional reconstruction schemes can be divided into active and passive schemes: the active three-dimensional reconstruction method mainly comprises a structured light method and a laser scanning method. The structured light method requires a main projection of structured light, and the current structured light product is Kinect of PrimeSense corporation in israel. Although the reconstruction precision is higher, the reconstruction precision can reach millimeter level, and the reasoning and operation speed is high. However, the active three-dimensional reconstruction method often has the problems of high cost, complex operation and the like, and for the precious animals such as pandas, the active three-dimensional reconstruction method adopting 3D laser scanning and the like is difficult and impractical. The passive three-dimensional reconstruction mode only uses a camera to acquire a three-dimensional scene to obtain a projected two-dimensional image, and three-dimensional reconstruction is realized according to image information. According to the method, the target object can be subjected to three-dimensional reconstruction only by using the two-dimensional image without adopting professional equipment. In passive three-dimensional reconstruction, for the case where the target object is an Animal, a Skinned Multi-Animal Linear model (SMAL) proposed by silvera Zuffi et al in 2017 abstracts the Animal three-dimensional model into parameters for expression. The SMAL model is a parameterized animal three-dimensional model, and the deformation of the three-dimensional model can be completed by taking shape parameters, posture parameters and translation parameters as input.
At present, no people or team researches the three-dimensional reconstruction of the pandas, but the three-dimensional reconstruction of the pandas is very urgent in the actual panda research field. The common data form of the pandas is a two-dimensional image, and meanwhile, in combination with the current research situation of animal three-dimensional reconstruction, learning a three-dimensional model from an image or a video frame is a research mode which is simpler and has obvious effect in realizing animal three-dimensional reconstruction. Therefore, the method for carrying out three-dimensional reconstruction on the pandas by adopting the image is the preferred scheme in the field of the existing panda three-dimensional reconstruction.
Disclosure of Invention
Aiming at the defects in the prior art, the panda three-dimensional model reconstruction method based on energy optimization provided by the invention reconstructs a panda three-dimensional model from a two-dimensional image by combining the prior information of the panda SMAL model, thereby filling the blank of the panda three-dimensional reconstruction field.
In order to achieve the above purpose, the invention adopts the technical scheme that:
the scheme provides a panda three-dimensional model reconstruction method based on energy optimization, which comprises the following steps:
s1, inputting a panda two-dimensional image and a panda parameterized three-dimensional model SMAL;
s2, extracting key points and outline silhouette segmentation maps of the panda two-dimensional image, and setting a corresponding relation between the key points and the top points of the panda parameterized three-dimensional model SMAL;
s3, updating the shape parameters, posture parameters and translation parameters of the panda parameterized three-dimensional model SMAL to obtain a new panda three-dimensional model, and carrying out perspective projection on the new panda three-dimensional model to obtain corresponding projection key points and projection areas;
s4, matching the projection key points with the key points of the panda two-dimensional image, matching the outline silhouette segmentation graph with the projection area, and constructing an energy optimization function according to two matching results and the corresponding relation;
s5, judging whether the energy optimization function is smaller than a threshold value, if so, entering a step S6, otherwise, minimizing the energy optimization function by adopting a gradient descent method, and returning to the step S3;
and S6, outputting the estimated shape parameters, posture parameters and translation parameters, combining the shape parameters, posture parameters and translation parameters with the panda parameterized three-dimensional model SMAL, and reconstructing the deformed panda three-dimensional model.
Further, the energy optimization function in step S4 includes a keypoint reprojection minimization function, a contour reprojection minimization function, a camera parameter penalty term, a shape penalty term, and a pose penalty term.
Still further, the expression of the energy optimization function is as follows:
E(β,r,t)＝Ekp(β,r,t)+Esil(β,r,t)+Ecam(f)+Eshape(β)+Epose(r)
wherein E (-) represents an energy optimization function, Ekp(. represents a keypoint reprojection minimization function, Esil(. represents a contour reprojection minimization function, Ecam(. represents a camera parameter penalty term, Eshape(. represents a shape penalty term, Epose(. cndot.) represents a pose penalty term, β represents a shape parameter, r represents a pose parameter, t represents a translation parameter, and f represents a focal length.
Still further, the expression of the keypoint reprojection minimization function is as follows:
the expression of the keypoint reprojection minimization function is as follows:
Figure BDA0002597953860000041
wherein E iskp(. h) represents a key point reprojection minimization function, nk represents the number of key points in the two-dimensional image, ρGMRepresenting a robust error function, kjRepresenting the jth image keypoint, nH(j)Representing the number of sets of three-dimensional vertices, v (β, r, t), corresponding to the jth image keypointk,j,mRepresenting the mth three-dimensional vertex in the set of three-dimensional vertex corresponding to the jth image keypoint, f representing a focal distance, β representing a shape parameter, r representing a pose parameter, t representing a translation parameter, and ii (·) representing the projection of the mth three-dimensional vertex on the image plane of the focal distance f.
Still further, the expression of the contour re-projection minimization function is as follows:
Figure BDA0002597953860000042
Figure BDA0002597953860000043
wherein E issil(. cndot.) represents a contour reprojection minimization function, β represents a shape parameter, r represents an attitude parameter, t represents a translation parameter,
Figure BDA0002597953860000044
representing the distance transformation, S representing the true segmentation contour, x representing the projection point of the three-dimensional model vertex through perspective projection, rhoGMA robust error function is represented as a function of,
Figure BDA0002597953860000045
a three-dimensional model outline diagram showing the perspective projection of the parameterized three-dimensional model SMAL of the panda at the focal length f after deformation, M shows the parameterized three-dimensional model SMAL of the panda, and pi (phi) shows the mth three-dimensional vertex at the focal length ff on the image plane of the image plane,
Figure BDA0002597953860000046
and representing projection points obtained by perspective projection of the vertexes of the deformed three-dimensional model.
The invention has the beneficial effects that:
(1) the method utilizes an energy optimization mode, combines the prior information of the panda SMAL model, can directly reconstruct a three-dimensional model of the panda from a two-dimensional image, fills the blank of the field of the three-dimensional reconstruction of the panda, can change a body ruler measuring mode of the panda from a contact type to a non-contact type, reduces the cost, saves resources and simultaneously improves the efficiency, and simultaneously, people can more intuitively and comprehensively know and know the panda in a flexible and various form through rapid three-dimensional reconstruction, three-dimensional display and the like, thereby having more interest and playing a certain role in further researching and protecting the panda.
(2) Compared with the traditional unrealistic mode of actively acquiring the panda three-dimensional data, the method can reconstruct the panda model only through the two-dimensional image, and is simple, high in reconstruction efficiency and strong in applicability.
(3) According to the method, the key points obtained by the deformed three-dimensional model perspective projection are matched with the key points correctly labeled by the target image through the key point re-projection minimizing function, and the reconstructed three-dimensional model projection and the current image have similar appearance characteristics as far as possible.
(4) The invention can control the shape contour of the panda within a certain reasonable range through the contour reprojection function, and ensure that all vertexes in the panda three-dimensional model are subjected to perspective.
(5) In real life, the focal length parameter f is a parameter of the digital camera due to defects of the sensor, errors in the calibration of the camera, and distortion caused by the lens of the cameraxAnd fyOften have different values, and in order to reduce the actual error existing in the method, f is set by a penalty term of a camera parameter penalty termxAnd fyHave the same value.
(6) According to the method, the shape parameters in the energy optimization process are limited within a reasonable range through the shape penalty term.
Drawings
FIG. 1 is a flow chart of the method of the present invention.
Fig. 2 is a schematic diagram of two-dimensional image panda key points in the invention.
Fig. 3 is a schematic diagram of segmentation of a two-dimensional image panda contour in the 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.
Examples
Pandas can be found in Sichuan, Gansu and Shaanxi areas of China. The body length is about 1.5m to 1.8m, the tail length is 12cm to 15cm, and the body weight is 80kg to 125kg, and the animal is an ancient, rare and precious special product animal in China. The study on the panda three-dimensional reconstruction algorithm can truly reflect the space morphological structure of the panda, propagate and display the panda model in the most intuitive form, and further help researchers to study and protect the panda. The invention provides a panda three-dimensional model recovery method based on an energy optimization mode, which is as shown in figure 1 and comprises the following steps:
s1, inputting a panda two-dimensional image and a panda parameterized three-dimensional model SMAL;
s2, extracting key points and outline silhouette segmentation maps of the panda two-dimensional image, and setting a corresponding relation between the key points and the top points of the panda parameterized three-dimensional model SMAL;
in this embodiment, as shown in fig. 2 to 3, the key points and the contour silhouette segmentation map of the two-dimensional image are manually extracted, and the corresponding relationship between the key points and the three-dimensional vertices in the parameterized three-dimensional model SMAL of the panda is set. The specific method for manually extracting the key point and outline silhouette segmentation graph through the two-dimensional panda image comprises the following steps: firstly, preprocessing a two-dimensional image, zooming the image to ensure that the size of the image is not more than 480 pixels, and simultaneously, manually extracting two-dimensional key points K and segmenting the outline of a panda from the preprocessed image, wherein 28 related key points are used for representing the panda, as shown in table 1, table 1 is a two-dimensional image panda key point representation table which comprises 4 foot key points, 4 knee key points, 4 ankle joint key points, 2 shoulder key points, left and right eye key points, left and right nose key points, left and right cheek key points, 2 mouth key points, 2 left and right ear key points, and key points of the front and back of the tail, the neck and the chin. In this example, an image size of 480 × 480 pixels is used.
TABLE 1
Figure BDA0002597953860000061
Figure BDA0002597953860000071
S3, updating the shape parameters, posture parameters and translation parameters of the panda parameterized three-dimensional model SMAL to obtain a new panda three-dimensional model, and carrying out perspective projection on the new panda three-dimensional model to obtain corresponding projection key points and projection areas;
in this embodiment, the shape parameter β, the posture parameter r, and the translation parameter t of the panda SMAL model are updated, the deformation of the panda SMAL model is controlled, a new panda three-dimensional model is obtained, and the deformed model is subjected to perspective projection to obtain a corresponding 2D plane projection diagram, which includes key points and a projection area.
S4, matching the projection key points with the key points of the panda two-dimensional image, matching the outline silhouette segmentation graph with the projection area, and constructing an energy optimization function according to two matching results and the corresponding relation;
in this embodiment, the panda three-dimensional model is subjected to perspective projection to obtain a projective relationMatching the key points with the key points marked by the two-dimensional image to ensure that all the projected key points fall in the target area of the segmentation graph to perform image fitting so as to construct an energy optimization function, wherein the energy optimization function comprises a key point re-projection minimization function EkpContour reprojection minimization function EsilPenalty term for camera parameters EcamShape penalty term EsapeAttitude penalty term Epose。
In this embodiment, the minimization function E is re-projected through the key pointskpAnd matching the key points obtained by the deformed three-dimensional model perspective projection with the key points correctly labeled by the target image, and ensuring that the reconstructed three-dimensional model projection has similar appearance characteristics with the current image as far as possible. The expression of the keypoint reprojection minimization function is as follows:
Figure BDA0002597953860000081
wherein E iskp(. h) represents a key point reprojection minimization function, nk represents the number of key points in the two-dimensional image, ρGMRepresenting a robust error function, kjRepresenting the jth image keypoint, nH(j)Representing the number of sets of three-dimensional vertices, v (β, r, t), corresponding to the jth image keypointk,j,mRepresenting the mth three-dimensional vertex in a set of three-dimensional vertex sets corresponding to the jth image keypoint, f representing a focal length, v (beta, r, t) representing a set of three-dimensional mesh vertex coordinates in the giant panda parameterized three-dimensional model SMAL, beta representing a shape parameter, r representing a posture parameter, t representing a translation parameter, and pi (·) representing the projection of the mth three-dimensional vertex on an image plane of the focal length f.
In this embodiment, the shape and contour of the panda can be controlled within a certain reasonable range by the contour re-projection minimizing function, and it is ensured that projection points obtained by perspective projection of all vertexes in the three-dimensional model of the panda are located as far as possible inside the contour segmentation map. The expression of the contour reprojection minimization function is as follows:
Figure BDA0002597953860000082
Figure BDA0002597953860000083
wherein E issil(. cndot.) represents a contour reprojection minimization function, β represents a shape parameter, r represents an attitude parameter, t represents a translation parameter,
Figure BDA0002597953860000084
representing the distance transformation, S representing the true segmentation contour, x representing the projection point of the three-dimensional model vertex through perspective projection, rhoGMA robust error function is represented as a function of,
Figure BDA0002597953860000085
a three-dimensional model outline diagram showing perspective projection of the parameterized three-dimensional model SMAL of the pandas on the focal length f after deformation, M shows the parameterized three-dimensional model SMAL of the pandas, pi (·) shows projection of the mth three-dimensional vertex on an image plane of the focal length f,
Figure BDA0002597953860000091
and representing projection points obtained by perspective projection of the vertexes of the deformed three-dimensional model.
In the embodiment, in real life, the focal length parameter f is caused by defects of a digital camera sensor, errors in camera calibration, distortion caused by a camera lens, and the likexAnd fyOften with different values. In order to reduce the actual error, the penalty item of the camera parameter is set to be fxAnd fyHave the same value.
In this embodiment, the shape penalty term limits the shape parameters in the energy optimization process within a reasonable range. In order to conform the shape parameter estimate to the prior distribution of panda shape parameters, the shape penalty term uses the mean and covariance matrices in the shape sample, EsapeDefined as the square of its mahalanobis distance.
In this embodiment, the attitude penalty term uses the mean and covariance matrices in the attitude samples,will EposeDefined as the square of its mahalanobis distance.
The total energy optimization function is:
E(β,r,t)＝Ekp(β,r,t)+Esil(β,r,t)+Ecam(f)+Eshape(β)+Epose(r)
wherein E (-) represents an energy optimization function, Ekp(. represents a keypoint reprojection minimization function, Esil(. represents a contour reprojection minimization function, Ecam(. represents a camera parameter penalty term, Eshape(. represents a shape penalty term, Epose(. cndot.) represents a pose penalty term, β represents a shape parameter, r represents a pose parameter, t represents a translation parameter, and f represents a focal length.
S5, judging whether the energy optimization function is smaller than a threshold value, if so, entering a step S6, otherwise, minimizing the energy optimization function by adopting a gradient descent method, and returning to the step S3;
and S6, outputting the estimated shape parameters, posture parameters and translation parameters, combining the shape parameters, posture parameters and translation parameters with the panda parameterized three-dimensional model SMAL, and reconstructing the deformed panda three-dimensional model.
The method utilizes an energy optimization mode and combines the prior information of the panda SMAL model, can directly reconstruct a three-dimensional model of the panda from a two-dimensional image, fills the blank of the field of the three-dimensional reconstruction of the panda, can change a body ruler measuring mode of the panda from a contact mode to a non-contact mode, reduces the cost, saves resources and simultaneously improves the efficiency, and simultaneously enables people to have more visual and comprehensive understanding and cognition on the panda in a flexible and various mode through rapid three-dimensional reconstruction, three-dimensional display and the like, thereby having more interest and playing a certain role in further researching and protecting the panda.

Claims (5)

1. A panda three-dimensional model reconstruction method based on energy optimization is characterized by comprising the following steps:
s1, inputting a panda two-dimensional image and a panda parameterized three-dimensional model SMAL;
s2, extracting key points and outline silhouette segmentation maps of the panda two-dimensional image, and setting a corresponding relation between the key points and the top points of the panda parameterized three-dimensional model SMAL;
s3, updating the shape parameters, posture parameters and translation parameters of the panda parameterized three-dimensional model SMAL to obtain a new panda three-dimensional model, and carrying out perspective projection on the new panda three-dimensional model to obtain corresponding projection key points and projection areas;
s4, matching the projection key points with the key points of the panda two-dimensional image, matching the outline silhouette segmentation graph with the projection area, and constructing an energy optimization function according to two matching results and the corresponding relation;
s5, judging whether the energy optimization function is smaller than a threshold value, if so, entering a step S6, otherwise, minimizing the energy optimization function by adopting a gradient descent method, and returning to the step S3;
and S6, outputting the estimated shape parameters, posture parameters and translation parameters, combining the shape parameters, posture parameters and translation parameters with the panda parameterized three-dimensional model SMAL, and reconstructing the deformed panda three-dimensional model.
2. The energy-based optimized panda three-dimensional model reconstruction method according to claim 1, wherein the energy optimization function in step S4 includes a key point reprojection minimization function, a contour reprojection minimization function, a camera parameter penalty term, a shape penalty term and a pose penalty term.
3. The energy-optimization-based panda three-dimensional model reconstruction method according to claim 2, wherein the expression of the energy optimization function is as follows:
E(β,r,t)＝Ekp(β,r,t)+Esil(β,r,t)+Ecam(f)+Eshape(β)+Epose(r)
wherein E (-) represents an energy optimization function, Ekp(. represents a keypoint reprojection minimization function, Esil(. represents a contour reprojection minimization function，Ecam(. represents a camera parameter penalty term, Eshape(. represents a shape penalty term, Epose(. cndot.) represents a pose penalty term, β represents a shape parameter, r represents a pose parameter, t represents a translation parameter, and f represents a focal length.
4. The energy-optimization-based panda three-dimensional model reconstruction method according to claim 3, wherein the expression of the keypoint reprojection minimization function is as follows:
Figure FDA0002597953850000021
wherein E iskp(. h) represents a key point reprojection minimization function, nk represents the number of key points in the two-dimensional image, ρGMRepresenting a robust error function, kjRepresenting the jth image keypoint, nH(j)Representing the number of sets of three-dimensional vertices, v (β, r, t), corresponding to the jth image keypointk,j,mRepresenting the mth three-dimensional vertex in the set of three-dimensional vertex corresponding to the jth image keypoint, f representing a focal distance, β representing a shape parameter, r representing a pose parameter, t representing a translation parameter, and ii (·) representing the projection of the mth three-dimensional vertex on the image plane of the focal distance f.
5. The energy-optimized panda three-dimensional model reconstruction method according to claim 3, wherein the expression of the contour re-projection minimizing function is as follows:
Figure FDA0002597953850000022
Figure FDA0002597953850000023
wherein E issil(. cndot.) represents a contour reprojection minimization function, β represents a shape parameter, r represents an attitude parameter, t represents a translation parameter,
Figure FDA0002597953850000024
representing the distance transformation, S representing the true segmentation contour, x representing the projection point of the three-dimensional model vertex through perspective projection, rhoGMA robust error function is represented as a function of,
Figure FDA0002597953850000025
a three-dimensional model outline diagram showing perspective projection of the parameterized three-dimensional model SMAL of the pandas on the focal length f after deformation, M shows the parameterized three-dimensional model SMAL of the pandas, pi (·) shows projection of the mth three-dimensional vertex on an image plane of the focal length f,
Figure FDA0002597953850000026
and representing projection points obtained by perspective projection of the vertexes of the deformed three-dimensional model.
CN202010715308.XA 2020-07-23 2020-07-23 Energy optimization-based panda three-dimensional model reconstruction method Withdrawn CN111862313A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010715308.XA CN111862313A (en) 2020-07-23 2020-07-23 Energy optimization-based panda three-dimensional model reconstruction method

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010715308.XA CN111862313A (en) 2020-07-23 2020-07-23 Energy optimization-based panda three-dimensional model reconstruction method

Publications (1)

Publication Number Publication Date
CN111862313A true CN111862313A (en) 2020-10-30

Family

ID=72949221

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010715308.XA Withdrawn CN111862313A (en) 2020-07-23 2020-07-23 Energy optimization-based panda three-dimensional model reconstruction method

Country Status (1)

Country Link
CN (1) CN111862313A (en)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112750199A (en) * 2021-01-13 2021-05-04 电子科技大学 Energy optimization-based panda three-dimensional model reconstruction method
CN113870267A (en) * 2021-12-03 2021-12-31 深圳市奥盛通科技有限公司 Defect detection method, defect detection device, computer equipment and readable storage medium
CN117409973A (en) * 2023-12-13 2024-01-16 成都大熊猫繁育研究基地 Panda health assessment method and system based on family data

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112750199A (en) * 2021-01-13 2021-05-04 电子科技大学 Energy optimization-based panda three-dimensional model reconstruction method
CN113870267A (en) * 2021-12-03 2021-12-31 深圳市奥盛通科技有限公司 Defect detection method, defect detection device, computer equipment and readable storage medium
CN113870267B (en) * 2021-12-03 2022-03-22 深圳市奥盛通科技有限公司 Defect detection method, defect detection device, computer equipment and readable storage medium
CN117409973A (en) * 2023-12-13 2024-01-16 成都大熊猫繁育研究基地 Panda health assessment method and system based on family data
CN117409973B (en) * 2023-12-13 2024-05-17 成都大熊猫繁育研究基地 Panda health assessment method and system based on family data

Similar Documents

Publication Publication Date Title
CN109544677B (en) Indoor scene main structure reconstruction method and system based on depth image key frame
US10846903B2 (en) Single shot capture to animated VR avatar
US11514642B2 (en) Method and apparatus for generating two-dimensional image data describing a three-dimensional image
CN107403463B (en) Human body representation with non-rigid parts in an imaging system
CN112750199A (en) Energy optimization-based panda three-dimensional model reconstruction method
CN112613609B (en) Nerve radiation field enhancement method based on joint pose optimization
CN111862313A (en) Energy optimization-based panda three-dimensional model reconstruction method
JP7015152B2 (en) Processing equipment, methods and programs related to key point data
CN111932678B (en) Multi-view real-time human motion, gesture, expression and texture reconstruction system
CN113366491B (en) Eyeball tracking method, device and storage medium
CN104537705A (en) Augmented reality based mobile platform three-dimensional biomolecule display system and method
Yin et al. Virtual reconstruction method of regional 3D image based on visual transmission effect
CN112509110A (en) Automatic image data set acquisition and labeling framework for land confrontation intelligent agent
Hwang et al. 3D face modeling using the multi-deformable method
Scataglini et al. Overview of software and file exchange formats in 3D and 4D body shape scanning
CN116342621A (en) Geometric parameter identification method and system based on three-dimensional reconstruction of space moving target
Zhang et al. CuFusion: Accurate real-time camera tracking and volumetric scene reconstruction with a cuboid
Ju et al. Individualising Human Animation Models.
CN116616812A (en) NeRF positioning-based ultrasonic autonomous navigation method
Li et al. Detail 3D face reconstruction based on 3DMM and displacement map
Khan et al. Skeleton based human action recognition using a structured-tree neural network
JPH0273471A (en) Estimating method for three-dimensional form
Wang et al. Markerless body motion capturing for 3d character animation based on multi-view cameras
Rhee et al. Soft-tissue deformation for in vivo volume animation
Kim et al. Temporal Estimation of Non-Rigid Dynamic Human Point Cloud Sequence Using 3D Skeleton-Based Deformation for Compression

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
WW01 Invention patent application withdrawn after publication

Application publication date: 20201030

WW01 Invention patent application withdrawn after publication