CN113284249A - Multi-view three-dimensional human body reconstruction method and system based on graph neural network - Google Patents
Multi-view three-dimensional human body reconstruction method and system based on graph neural network Download PDFInfo
- Publication number
- CN113284249A CN113284249A CN202110652098.9A CN202110652098A CN113284249A CN 113284249 A CN113284249 A CN 113284249A CN 202110652098 A CN202110652098 A CN 202110652098A CN 113284249 A CN113284249 A CN 113284249A
- Authority
- CN
- China
- Prior art keywords
- human body
- neural network
- model
- graph
- 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.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 43
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000013507 mapping Methods 0.000 claims abstract description 7
- 238000005516 engineering process Methods 0.000 claims abstract description 4
- 210000000988 bone and bone Anatomy 0.000 claims description 15
- 230000006870 function Effects 0.000 claims description 5
- 238000012549 training Methods 0.000 claims description 5
- 238000012935 Averaging Methods 0.000 claims description 4
- 230000037237 body shape Effects 0.000 claims description 4
- 230000004913 activation Effects 0.000 claims description 3
- 238000004364 calculation method Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 abstract description 3
- 238000012545 processing Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000009499 grossing Methods 0.000 description 1
- 230000001939 inductive effect Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
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
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2200/00—Indexing scheme for image data processing or generation, in general
- G06T2200/04—Indexing scheme for image data processing or generation, in general involving 3D image data
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Processing Or Creating Images (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a multi-view three-dimensional human body reconstruction method and a system based on a graph neural network, wherein the method comprises the steps of collecting human body images after a human body is placed in the center of an annular camera; acquiring a plurality of images, selecting a reference image from the images, and recording the rest images as source images; extracting the features of the image, inputting the features into an attention network of the image so as to obtain parameters of a parameterized human body model, and reconstructing a human body three-dimensional model in the output of the parameterized human body model by using the output parameters; and (3) re-projecting the obtained parameterized human body model onto each image, aligning, and projecting color and texture information in the image onto a triangulated vertex of the model by using a computer graphics technology to finish texture mapping. The method utilizes the graph neural network to complete the extraction of key points and skin vertex information in the parameterized human body model, and can realize fine human body three-dimensional reconstruction work.
Description
Technical Field
The invention relates to the technical field of computer vision, in particular to a multi-view three-dimensional human body reconstruction method and system based on a graph neural network.
Background
The rise and application of neural networks has been very successful in the past years to advance the field of computer vision, and many machine learning tasks that had relied heavily on manual feature extraction have now been altered by various deep learning neural networks. Recently, people are more and more interested in the expansion of a deep learning method on a graph, and researchers design a neural network structure aiming at graph data, namely a graph neural network, on the idea of using a convolutional network, a cyclic network and a deep automatic encoder for reference.
On the other hand, human three-dimensional reconstruction has been a research focus in the field of computer vision. The three-dimensional reconstruction method comprises two main reconstruction modes, namely, three-dimensional reconstruction is performed by using a single image with depth information, but the single image method has less information amount and cannot completely restore a three-dimensional scene, and the other method is three-dimensional reconstruction after stereo matching is performed on images shot from multiple visual angles, and various methods for extracting depth information based on depth learning already exist.
How to introduce the neural network into the human body three-dimensional reconstruction so as to realize more accurate human body three-dimensional reconstruction is a technical problem which is urgently needed to be solved at present.
Disclosure of Invention
In order to solve the technical problem of how to realize more accurate human body three-dimensional reconstruction, the invention provides a multi-view three-dimensional human body reconstruction method and system based on a graph neural network.
Therefore, the multi-view three-dimensional human body reconstruction method based on the graph neural network provided by the invention comprises the following steps:
a2, when a human body is placed in the center of the annular camera, acquiring a human body image;
a3, acquiring a plurality of images, selecting a reference image from the images, and recording the rest images as source images;
a4, constraining each reference image and the source images at the two sides of the reference image in the middle of the adjacent reference image by using polar line constraint conditions in the multi-view geometry, searching to obtain a plurality of groups of skeleton point coordinates of which the number and the positions accord with the definition of a parameterized human body model, averaging the plurality of groups of skeleton point coordinates to obtain a final group of joint coordinates, extracting characteristic points on the surface of the human body to generate human body shape sparse point clouds serving as surface contour mesh preset vertexes of the parameterized human body model;
a5, selecting partial point clouds around the skeleton points, associating the skeleton points with the selected point clouds to form natural image data, inputting image attention neural network training to obtain the posture information parameters and body type information parameters of a parameterized human body model, and reconstructing a human body three-dimensional model in the output of the parameterized human body model by utilizing output parameters;
and A6, re-projecting the obtained parameterized human body model onto each image, aligning, and projecting color and texture information in the image onto a triangulated vertex of the model by using a computer graphics technology to complete texture mapping.
Further, before the step a2, a step a1 is further included, and the step a1 specifically includes erecting a plurality of cameras in a circular arrangement around the human body capture area, and calibrating the plurality of cameras.
Further, calibrating the plurality of cameras specifically includes performing internal-participation external reference calibration on the cameras by a black-and-white chess-card-grid calibration method.
Further, in the step a4, feature points of the human body surface are extracted by using a scale-invariant feature transform algorithm.
Further, in the step a5, a part of the point cloud around the bone point is extracted by using a K-nearest neighbor method.
Further, in the step a5, the key point information of the posture information is obtained through the graph automatic coding neural network, and then the key point information and the triangulated vertex composition graph are input into the graph attention neural network, and the body type information in the parameterized human body model is obtained through learning.
Further, in the step a5, the graph attention neural network is implemented by a graph attention layer stack, and the attention calculation coefficient of each graph attention layer is:
wherein alpha isijIs the attention coefficient, N, from the vertex j to the bone point iiAll neighboring vertices, W, representing bone points iijRepresents the weight coefficient between the vertex j and the bone point i, i.e. the L2 norm, a represents the weight vector, and LeakyReLU representsThe LeakyReLU activation function.
Further, in step a6, color information in the pixels of the video is extracted by using a computer graphics technique, and the extracted color information is re-projected onto the skin vertex of the parameterized human body model, so as to complete texture mapping.
Further, the parameterized human body model is selected as a skin multi-person linear model, the skin multi-person linear model comprises two parameters of posture information theta and body type information beta, the posture information theta is defined by 24 human body key points, and the body type information beta is 10 parameters for controlling the change of the shape of the human body and is expressed as a triangulated skin vertex forming the surface of the model.
The multi-view three-dimensional human body reconstruction system based on the graph neural network comprises a camera, a processor and a memory, wherein a program which can be operated by the processor is stored in the memory, and the processor can realize the multi-view three-dimensional human body reconstruction method based on the graph neural network by operating the program.
Compared with the prior art, the invention has the following beneficial effects:
in the process of applying the graph neural network to human body three-dimensional reconstruction, the graph neural network is utilized to complete the extraction of key points and skin vertex information in the parameterized human body model, and the fine human body three-dimensional reconstruction work can be realized.
In some embodiments of the invention, the following advantages are also provided:
the parameterized human body model is selected as a skin multi-person linear model, and the SMPL (skin multi-person linear) human body three-dimensional reconstruction model can be well imported into Unity, Maya and other application programs, so that the multi-view three-dimensional human body reconstruction method based on the graph neural network can be used for manufacturing animations, virtual reality and the like.
Drawings
Fig. 1 is a flow chart of a multi-view three-dimensional human body reconstruction method based on a graph neural network.
Detailed Description
In order to more clearly understand the technical features, objects, and effects of the present invention, specific embodiments of the present invention will now be described with reference to the accompanying drawings.
As shown in fig. 1, the multi-view three-dimensional human body reconstruction method based on the graph neural network provided by the embodiment of the invention includes the following steps:
a1, marking N cameras as 0,1,2 … … N-1, erecting the N cameras around a human body capturing area in an annular arrangement mode, and calibrating the N cameras by using a traditional camera calibration method to obtain internal reference and external reference information of the cameras. In the invention, 15 industrial cameras can be used and respectively marked as 0,1 and 2 … … 14, the 15 industrial cameras are placed on an experimental site, the height of the cameras is about 1.2 meters away from the ground, the horizontal distance between every two adjacent cameras is about 0.8-0.9 meter, and the cameras are circularly arranged according to the marked sequence. Or a circle can be drawn by taking the center point of the human body capturing area as the center of the circle and taking about 2 meters as the radius, and the camera can be erected at the equal division point after the circle 15 is divided equally. Specifically, in practical application, the number of cameras, the erection height, the shooting range size and the like can be set according to the requirements of experimenters. Calibrating the N cameras to need calibrating internal references and external references of the cameras, firstly printing a black-and-white checkerboard to be attached to a plane as a calibration object, respectively shooting 15 different pictures of the checkerboard by adjusting the direction of the calibration object for each Camera by using the Camera to be calibrated, preferably, guiding the 15 pictures into a Camera calibration tool box in MATLAB, setting the size of the checkerboard, extracting angular points and calculating five internal references and six external references of the cameras.
And A2, when the human body is placed at the center of the annular camera, acquiring the human body image. When the human body is arranged at the position of a circle center surrounded by the centers of the annular cameras, the calibrated 15 cameras are used for shooting the human body simultaneously to finish image acquisition.
And A3, processing the acquired N images to convert the acquired N images into RGB type images, selecting three images acquired by three cameras with the labels of 0, N/3 and 2N/3 as three reference images respectively in the images, and marking the rest images as source images. Specifically, the function in the OpenCV open source library is used to convert the image processing into RGB images, and the human body images acquired by the cameras 0,5 and 10 are respectively recorded as reference images, and the human body images acquired by the other cameras are recorded as source images.
A4, constraining each reference image and the source images at two sides of each reference image in the middle of the adjacent reference images by using polar line constraint conditions in multi-view geometry, searching to obtain three groups of skeleton point coordinates of which the number and the positions conform to the SMPL (skin multi-person linear) model definition, averaging the three groups of skeleton point coordinates to obtain a final group of joint coordinates, extracting feature points on the surface of a human body by using a SIFT algorithm to generate human body shape sparse point clouds serving as surface contour mesh preset vertexes of the SMPL human body three-dimensional reconstruction model. Specifically, corresponding skeleton point positions are marked on reference images acquired by cameras with the labels of 0,5 and 10 according to 24 skeleton point positions in an SMPL (simple Markov chain) human body model, the three-dimensional space position coordinates of the skeleton points are searched for the No. 0 image and the images with the labels of 11, 12, 13, 14, 1,2, 3 and 4 by using an epipolar constraint condition, and the three-dimensional space position coordinates of the skeleton points are searched for the No. 5 and No. 10 images and 8 source images adjacent to the two sides of the No. 5 and No. 10 images in the same manner. And averaging the three groups of obtained skeleton point coordinates to obtain final skeleton point coordinates. Preferably, a SIFT algorithm (scale invariant feature transform algorithm) is used to obtain a sparse point cloud of the human body surface contour in the image, and the point cloud can be used as a preset vertex of the skin surface of the SMPL model. Specifically, the SMPL human body three-dimensional reconstruction model comprises two parameters of posture information theta and body type information beta, wherein the posture information theta is defined by 24 human body key points, and the body type information beta is 10 parameters for controlling the change of the human body shape and is expressed as a triangularized skin vertex forming the surface of the model. The SMPL model may be imported into a graphics program such as Unity.
A5, selecting partial point clouds around the skeleton points by a K neighbor method, associating the skeleton points with the selected point clouds to form natural graph data, inputting GAT graph attention neural network training to obtain attitude information parameters and body type information parameters of the SMPL model, and reconstructing a human body three-dimensional model in SMPL output by using the output parameters. Preferably, a regression matrix is used to convert the three-dimensional coordinate representation of the bone points into a representation in the SMPL model, each bone point vector containing information on three angles of limb rotation (yaw, roll, pitch), and in particular, the regression matrix can be trained using an open source data set such as sureral or human3.6 m. Further, not less than 50 point cloud positions are selected around each bone point by using a K-nearest neighbor method, the bone points and the selected point clouds are combined into natural map data, and then the natural map data are trained by using a GAT (goal-oriented) map attention neural network. The GAT graph attention network is realized by stacking graph attention layers, and the attention calculation coefficient of each graph attention layer is as follows:
wherein alpha isijIs the attention coefficient, N, from the vertex j to the bone point iiAll neighboring vertices, W, representing bone points iijAnd a represents a weight coefficient between the vertex j and the bone point i, namely an L2 norm, and a represents a weight vector, and finally, after normalization processing is carried out by using a softmax function, nonlinearity is provided by using an LeakyReLU activation function. The advantage of using this structure is that the neighbor pairs of skeleton points and vertices can be computed in parallel, thus having higher computational efficiency, and the structure can process nodes of different degrees and assign corresponding weights to the neighbors. In addition, the GAT graph can realize training without knowing the structure of the whole graph, and SMPL human model parameters can be obtained through fast and accurate inductive learning. And obtaining posture parameters and body type parameters of the SMPL model after GAT graph attention network training of graph data formed by the skeletal points and the human body contour vertexes, and inputting the posture parameters and the body type parameters into the SMPL model to obtain a final SMPL human body model. Specifically, key point information of the posture information is obtained through a GAE graph automatic coding neural network, then the key point information and the triangularization vertex are combined into a graph to be input into a GAT graph attention network, and body type information in the SMPL model is obtained through learning.
And A6, re-projecting the obtained SMPL model on each image for alignment, and projecting the color and texture information in the image on the triangulated vertex of the model by using the computer graphics technology to complete texture mapping. And B, importing the SMPL model obtained in the step A5 into a Unity program, aligning all images with the SMPL model in a ring arrangement by combining internal and external parameters calibrated by a camera, re-projecting the SMPL model onto each image to acquire pixel color information of corresponding triangular vertexes, performing Gaussian smoothing noise reduction processing on the pixel information of the vertexes of the overlapped part, and finally completing the texture mapping process of the model.
The multi-view three-dimensional human body reconstruction system based on the graph neural network comprises a camera, a processor and a memory, wherein a program which can be operated by the processor is stored in the memory, and the processor can realize the multi-view three-dimensional human body reconstruction method based on the graph neural network by operating the program.
The method for applying the graph neural network to the human body three-dimensional reconstruction is different from the existing method for obtaining the depth map by using the convolutional neural network, and because the SMPL human body three-dimensional reconstruction model can be well imported into the Unity, Maya and other application programs, the method for deducing the SMPL human body three-dimensional reconstruction model parameters from a plurality of human body images by using the graph neural network can realize more accurate human body three-dimensional reconstruction and can be used for making animation, virtual reality and the like.
The above disclosure is only for the purpose of illustrating the preferred embodiments of the present invention, and it should not be understood that the scope of the present invention is limited thereby. It should be noted that those skilled in the art should recognize that they may make equivalent changes in the design and construction of the invention without departing from the spirit and scope of the invention.
Claims (10)
1. A multi-view three-dimensional human body reconstruction method based on a graph neural network is characterized by comprising the following steps:
a2, when a human body is placed in the center of the annular camera, acquiring a human body image;
a3, acquiring a plurality of images, selecting a reference image from the images, and recording the rest images as source images;
a4, constraining each reference image and the source images of two sides of each reference image positioned in the middle of the adjacent reference images by using polar line constraint conditions in multi-view geometry, searching to obtain a plurality of groups of skeleton point coordinates of which the number and the positions accord with the definition of a parameterized human body model, averaging the plurality of groups of skeleton point coordinates to obtain a final group of joint coordinates, extracting characteristic points on the surface of the human body to generate human body shape sparse point clouds serving as surface contour mesh preset vertexes of the parameterized human body model;
a5, selecting partial point clouds around the skeleton points, associating the skeleton points with the selected point clouds to form natural image data, inputting image attention neural network training to obtain the posture information parameters and body type information parameters of a parameterized human body model, and reconstructing a human body three-dimensional model in the output of the parameterized human body model by utilizing output parameters;
and A6, re-projecting the obtained parameterized human body model onto each image, aligning, and projecting color and texture information in the image onto a triangulated vertex of the model by using a computer graphics technology to complete texture mapping.
2. The method for multi-view three-dimensional human body reconstruction based on graph neural network of claim 1, further comprising a step a1 before the step a2, wherein the step a1 specifically comprises erecting a plurality of cameras in a circular arrangement around the human body capture area, and calibrating the plurality of cameras.
3. The method of claim 2, wherein the calibrating the plurality of cameras comprises internally participating in external reference calibration of the cameras by a black and white checkerboard calibration method.
4. The method for multi-view three-dimensional human body reconstruction based on graph neural network as claimed in claim 1, wherein in said step a4, feature points on the surface of human body are extracted by using a scale invariant feature transform algorithm.
5. The method for reconstructing a multi-view three-dimensional human body based on a graph neural network as claimed in claim 1, wherein in said step a5, a part of point cloud around said bone point is extracted by K-nearest neighbor method.
6. The method according to claim 1, wherein in step a5, the neural network is automatically encoded to obtain key point information of pose information, and the key point information and the triangulated vertex are combined into a graph and input to the graph attention neural network to learn the body type information in the parameterized human model.
7. The multi-view three-dimensional human body reconstruction method based on graph neural network as claimed in claim 1, wherein in said step a5, said graph attention neural network is implemented by graph attention layer stack, and the attention calculation coefficient of each graph attention layer is:
wherein alpha isijIs the attention coefficient, N, from the vertex j to the bone point iiAll neighboring vertices, W, representing bone points iijThe weight coefficient between the vertex j and the bone point i, i.e. the L2 norm, is represented, a represents the weight vector, and LeakyReLU represents the LeakyReLU activation function.
8. The method for multi-view three-dimensional human body reconstruction based on graph neural network as claimed in claim 1, wherein in said step a6, color information in the pixels of the video image is extracted by using computer graphics technique and is re-projected onto the skin vertex of the parameterized human body model to complete texture mapping.
9. The method for multi-view three-dimensional human body reconstruction based on graph neural network according to any one of claims 1-8, characterized in that the parameterized human body model is selected as a skinned multi-person linear model, the skinned multi-person linear model comprises two parameters of attitude information θ and body type information β, the attitude information θ is defined by 24 body key points, and the body type information β is 10 parameters for controlling the change of the shape of the human body and represents a triangulated skinned vertex forming the surface of the model.
10. A multi-view three-dimensional human body reconstruction system based on a graph neural network, which comprises a camera, a processor and a memory, wherein the memory stores a program for the processor to run, and the processor can realize the multi-view three-dimensional human body reconstruction method based on the graph neural network according to any one of claims 1 to 9 by running the program.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110652098.9A CN113284249B (en) | 2021-06-11 | 2021-06-11 | Multi-view three-dimensional human body reconstruction method and system based on graph neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110652098.9A CN113284249B (en) | 2021-06-11 | 2021-06-11 | Multi-view three-dimensional human body reconstruction method and system based on graph neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN113284249A true CN113284249A (en) | 2021-08-20 |
CN113284249B CN113284249B (en) | 2022-05-10 |
Family
ID=77284338
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110652098.9A Active CN113284249B (en) | 2021-06-11 | 2021-06-11 | Multi-view three-dimensional human body reconstruction method and system based on graph neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN113284249B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114998515A (en) * | 2022-05-19 | 2022-09-02 | 大连理工大学 | 3D human body self-supervision reconstruction method based on multi-view images |
CN116109522A (en) * | 2023-04-10 | 2023-05-12 | 北京飞渡科技股份有限公司 | Contour correction method, device, medium and equipment based on graph neural network |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108053469A (en) * | 2017-12-26 | 2018-05-18 | 清华大学 | Complicated dynamic scene human body three-dimensional method for reconstructing and device under various visual angles camera |
CN110599540A (en) * | 2019-08-05 | 2019-12-20 | 清华大学 | Real-time three-dimensional human body shape and posture reconstruction method and device under multi-viewpoint camera |
US20200058137A1 (en) * | 2015-06-24 | 2020-02-20 | Sergi PUJADES | Skinned Multi-Person Linear Model |
CN111627101A (en) * | 2020-05-22 | 2020-09-04 | 北京工业大学 | Three-dimensional human body reconstruction method based on graph convolution |
CN111932678A (en) * | 2020-08-13 | 2020-11-13 | 北京未澜科技有限公司 | Multi-view real-time human motion, gesture, expression and texture reconstruction system |
-
2021
- 2021-06-11 CN CN202110652098.9A patent/CN113284249B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20200058137A1 (en) * | 2015-06-24 | 2020-02-20 | Sergi PUJADES | Skinned Multi-Person Linear Model |
CN108053469A (en) * | 2017-12-26 | 2018-05-18 | 清华大学 | Complicated dynamic scene human body three-dimensional method for reconstructing and device under various visual angles camera |
CN110599540A (en) * | 2019-08-05 | 2019-12-20 | 清华大学 | Real-time three-dimensional human body shape and posture reconstruction method and device under multi-viewpoint camera |
CN111627101A (en) * | 2020-05-22 | 2020-09-04 | 北京工业大学 | Three-dimensional human body reconstruction method based on graph convolution |
CN111932678A (en) * | 2020-08-13 | 2020-11-13 | 北京未澜科技有限公司 | Multi-view real-time human motion, gesture, expression and texture reconstruction system |
Non-Patent Citations (8)
Title |
---|
HAOSONG YUE ET AL.: "3D body reconstruction using graph optimization", 《IEEE》 * |
KOLOTOUROS N ET AL.: "Learning to reconstruct 3D human pose and shape via model-fitting in the loop", 《ARXIV》 * |
LASSNER, C. ET AL.: "Unite the people: closing the loop between 3D and 2D human representations", 《IEEE CVPR》 * |
张皓若: "基于SMPL模型的三维人体建模及其应用研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
李健等: "结合形变模型的人体姿态估计优化算法", 《西安电子科技大学学报》 * |
李萌民: "基于多视角图神经网络的三维模型检索算法研究", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
汪磊 等: "基于图卷积神经网络的人体三维重建", 《广播电视网络》 * |
陈佳宇等: "基于二值图像的三维人体模型重建", 《毛纺科技》 * |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114998515A (en) * | 2022-05-19 | 2022-09-02 | 大连理工大学 | 3D human body self-supervision reconstruction method based on multi-view images |
CN116109522A (en) * | 2023-04-10 | 2023-05-12 | 北京飞渡科技股份有限公司 | Contour correction method, device, medium and equipment based on graph neural network |
Also Published As
Publication number | Publication date |
---|---|
CN113284249B (en) | 2022-05-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110458939B (en) | Indoor scene modeling method based on visual angle generation | |
CN110443842B (en) | Depth map prediction method based on visual angle fusion | |
CN109360240B (en) | Small unmanned aerial vehicle positioning method based on binocular vision | |
CN110853075B (en) | Visual tracking positioning method based on dense point cloud and synthetic view | |
CN110189399B (en) | Indoor three-dimensional layout reconstruction method and system | |
CN106780592A (en) | Kinect depth reconstruction algorithms based on camera motion and image light and shade | |
Barazzetti et al. | True-orthophoto generation from UAV images: Implementation of a combined photogrammetric and computer vision approach | |
CN108428255A (en) | A kind of real-time three-dimensional method for reconstructing based on unmanned plane | |
CN111243093A (en) | Three-dimensional face grid generation method, device, equipment and storage medium | |
Leotta et al. | Urban semantic 3D reconstruction from multiview satellite imagery | |
CN111950477B (en) | Single-image three-dimensional face reconstruction method based on video supervision | |
CN113284249B (en) | Multi-view three-dimensional human body reconstruction method and system based on graph neural network | |
CN110633628B (en) | RGB image scene three-dimensional model reconstruction method based on artificial neural network | |
CN109816784B (en) | Method and system for three-dimensional reconstruction of human body and medium | |
CN114359509B (en) | Multi-view natural scene reconstruction method based on deep learning | |
CN113421328B (en) | Three-dimensional human body virtual reconstruction method and device | |
CN107369204B (en) | Method for recovering basic three-dimensional structure of scene from single photo | |
CN111998862B (en) | BNN-based dense binocular SLAM method | |
CN104794728A (en) | Method for reconstructing real-time three-dimensional face data with multiple images | |
CN108648194A (en) | Based on the segmentation of CAD model Three-dimensional target recognition and pose measuring method and device | |
CN116740288B (en) | Three-dimensional reconstruction method integrating laser radar and oblique photography | |
CN106023307A (en) | Three-dimensional model rapid reconstruction method and system based on field environment | |
Kang et al. | Competitive learning of facial fitting and synthesis using uv energy | |
CN104751451B (en) | Point off density cloud extracting method based on unmanned plane low latitude high resolution image | |
CN115393519A (en) | Three-dimensional reconstruction method based on infrared and visible light fusion image |
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 |