CN107169475B - A kind of face three-dimensional point cloud optimized treatment method based on kinect camera - Google Patents
A kind of face three-dimensional point cloud optimized treatment method based on kinect camera Download PDFInfo
- Publication number
- CN107169475B CN107169475B CN201710464550.2A CN201710464550A CN107169475B CN 107169475 B CN107169475 B CN 107169475B CN 201710464550 A CN201710464550 A CN 201710464550A CN 107169475 B CN107169475 B CN 107169475B
- Authority
- CN
- China
- Prior art keywords
- depth
- pixel
- value
- face
- rgb
- 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.)
- Expired - Fee Related
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
-
- 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
- G06T17/005—Tree description, e.g. octree, quadtree
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/161—Detection; Localisation; Normalisation
- G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
Abstract
The invention belongs to three-dimensional reconstruction fields, are related to a kind of face three-dimensional point cloud optimized treatment method based on kinect camera.Technical solution of the present invention is handled different from traditional point cloud to three dimensional form, the scheme of the invention is directly being handled depth information using the four-way image RGB-D of Kinect camera to obtain spots cloud optimization effect.Method of the invention carries out the optimization process flow of face three-dimensional point cloud using existing Kinect depth camera, avoids the expensive price of traditional high precision apparatus or the problem that general algorithm precision is not high；Inside the depth optimization the step of, when carrying out surface recovery using depth after obtained optimization, it does not take and general is directly merged using normal vector, because the fusion of normal vector, which solves surface, can generate offset error, it is solved here with the geometrical relationship of concentration gradient and normal vector, last depth is directly obtained, finally obtains the higher three-dimensional point cloud of quality using the variation of coordinate.
Description
Technical field
The invention belongs to three-dimensional reconstruction field, it is related at a kind of face three-dimensional point cloud optimization based on kinect camera
Reason method.
Background technique
Currently, three-dimensional reconstruction is the research hotspot of computer vision field, human face three-dimensional model is even more a comparison
Important direction is frequently utilized for Film Animation production, game, medical treatment etc. associated scenario.The method of three-dimensional face is obtained at present
Including three kinds: modeling by hand, instrument acquisition and the modeling based on image.Modeling is needed as earliest three-dimensional modeling means by hand
Professional person is wanted to carry out；It is typically to represent based on structure light and laser scanner in instrument acquisition, but in general,
The cost of instrument is relatively high；Modeling based on image is mainly obtained as a result, this several width images using certain algorithm
Mode is fairly simple, but is not as a result very accurate or robust due to the limitation of algorithm itself.
But with the appearance of depth camera, such as kinect, the depth of face can easily and be rapidly got,
It is exactly easily to obtain the three-dimensional point cloud of face, but the depth value noise that kinect camera obtains is bigger, puts making an uproar for cloud
Sound is also very greatly, to need to optimize processing to a cloud.Traditional spots cloud optimization mainly directly to the point cloud of three dimensional form into
Row processing, for example, usually filling a vacancy from the deuterogenic triangle gridding of cloud aiming at the problem that missing；Not for distribution
The method that equal problem is generally based on optimal projection is handled；For the method that the noise of cloud is mainly based upon projection,
The method of improved filtering and method based on probability statistics.Directly three-dimensional point is handled, either also from data volume
It is for result, the robustness of algorithm is all very big uncertain factor.
Summary of the invention
It is to be solved by this invention to be, aiming at above-mentioned conventional method there are problem, propose that one kind is not based on kinect
The face three-dimensional point cloud optimized treatment method of camera.
The technical scheme is that as shown in Figure 1, a kind of face three-dimensional point cloud optimization processing based on kinect camera
Method, which comprises the following steps:
S1, kinect camera calibration, the inside and outside parameter including depth camera and color camera；
S2, tested facial image: the camera in applying step S1 is obtained, obtains the cromogram and depth map of tested face,
And the cromogram of acquisition and depth map are subjected to registration process, i.e., it is the different depth map of pixel size and cromogram one is a pair of
Together, each depth value is made to have a corresponding rgb value；
S3, the image of acquisition is pre-processed, the missing point including depth map, the erroneous point of cromogram, and to depth
Carry out bilateral filtering processing；
S4, face is extracted from the cromogram of acquisition, and the face extracted is projected to the depth map the inside of alignment,
Obtain corresponding human face region in depth map；
S5, RGB and depth D are combined, depth optimization processing is carried out to human face region corresponding in D figure, is recycled
Depth after optimization is converted into three-dimensional point cloud by the internal reference of depth camera.
Above scheme is the total technical solution of the present invention, compared with the mode that tradition is directly handled three-dimensional point,
Method of the invention utilizes the four-way image RGB-D of Kinect camera, is directly handled depth information to obtain a cloud
Effect of optimization, therefore conventional method is compared, method optimal speed of the invention is very fast, while result is also more stable.
Further, the step S3 method particularly includes:
For the missing point of depth D figure, it is assumed that missing pixel point coordinate is (i, j), then just above and below the point
Left and right four direction scans for, and until finding the first nonzero value in each direction, then four values is averaged to come
Fill the missing values of the pixel.For the erroneous point of RGB figure, the value in three channels of RGB figure is judged respectively first, if three
Channel is simultaneously 0, decides that the point is erroneous point, then assigning three channels of the pixel is all 1.
Further, as shown in Fig. 2, the step S4 method particularly includes:
S41, the cromogram of acquisition is mapped:
RGB color is mapped to YCbCr space by following formula 1:
In YCbCr space, the Cr component of normal yellow about between 140 to 175, Cb component about 100 to
Between 120, meet this rule be recognized as face skin, certainly include neck and a part of limb skin of exposing etc., because
This present invention carries out prescreening using the substantially value range of CbCr component.
S42, face is tentatively extracted:
Using each pixel in the transformation space image that dimensional Gaussian model inspection S41 is obtained shown in following formula 2
Point is the probability of the colour of skin:
After the probability value for obtaining each pixel, binaryzation is carried out to the probability graph of acquisition, is as a result preliminary extract
Face result；
S43, the bianry image of acquisition is handled, extracts target face rectangular area.The method of processing includes shape
State student movement is calculated, filling rate, area ratio, and length-width ratio etc. judges that the factor is further processed bianry image.
Further, the step S5 method particularly includes:
S51, eigen decomposition is carried out to RGB image, is decomposed according to illumination model shown in following formula 3:
In formula 3,It is the intensity at image slices vegetarian refreshments (i, j),It is the light and shade value of pixel, ρ (i, j)
Indicate multiple reflection rate, β (i, j) indicates local illumination variation impact factor；To three parametric solutions,.We are false first
If β=0, ρ=1, we solve S；Then the S solved is brought into, still assumes β=0, solves ρ；It will finally solve
ρ and S bring into, find out last β.
The parameter obtained in S52, solution procedure S51, specifically:
S521, light and shade value is solved
The light and shade value of image certain point is expressed as to the linear forms of the normal vector of the point shown in formula 4:
Wherein,It is the normal vector of each pixel of image,It is first four spheric harmonic function coefficient,Figure
Each pixel as in may serve to solve light and shade value, and therefore, need exist for solving is that the least square coefficient of overdetermination is estimated
Meter problem, following formula 5:
S522, multiple reflection rate ρ (i, j) is solved:
It is solved using energy function shown in following formula 6:
In formula 6, N is the neighborhood of certain pixel,It is weight coefficient relevant to pixel intensity value,It is and initial depth
Relevant weight coefficient is spent, I is gray value, and subscript k is pixel number, it is expressed as formula 7:
In formula 7, z (i, j) is the depth value of pixel (i, j), and I (i, j) is the color-values of pixel (i, j), σcIt is to close
The coefficient of pixel intensity value, σ in Mr. Yu's pixel and its neighborhooddIt is about the discontinuous related coefficient of depth；
S523, part illumination variation impact factor β (i, j) is solved:
β (i, j) is obtained by following formula 8:
S53, the surface recovery that face is carried out according to the parameter that step S52 is obtained, are minimized using energy, by following public
Formula 9 solves:
In formula 9, z0It is image initial depth value (being obtained by kinect camera), Δ indicates its Laplace operator；△
It is the local derviation relationship of depth, is the expression for the normal vector of each pixel, there is relationship shown in following formula 10:
In formula 10, the subscript x and y of z respectively indicates the gradient in the direction x and the gradient in the direction y；
The method for solving of formula 9 is specifically, because the non-linear relation of normal vector and concentration gradient, produces nonlinear terms
Problem is described as follows nonlinear solution throughway:
S531, fixed nonlinear terms, i.e. denominator term when solving normal vector, by this denominator term in terms of previous
Result is calculated as given value and brings the calculating of normal vector this time into, shown in following formula 11:
S532, using obtained normal vector, update linear light and shine model terms, following formula 12:
S533, iterative solution, according to parameterAs f (zk-1)>f(zk) set up when, always iterate to calculate:
Until f (zk+1)>f(zk), stop iterative process, previous zkThe as obtained last knot of our institute's optimization processings
Fruit.
In above scheme, weight threshold τ=0.05,λρ=0.1, Depth data is not converted into instead of between the 0-255 of grayscale image, true
Depth value, as unit of mm, value range is all converted between 0-2000mm.
Further, following formula 13:
In formula 13, (i, j) is the pixel in image, [x(i,j),y(i,j),z(i,j)] it is that pixel (i, j) to be solved is three-dimensional
Coordinate, zk(i, j) is its depth value, K-1It is inverting for depth camera internal reference matrix, according to transformation above, so that it may obtain three
Tie up point cloud data.
Further, further includes:
S6, the processing of rough error point is removed to the three-dimensional point cloud of acquisition:
Using the Octree module in the library PCL, the removal of rough error point is carried out to the three-dimensional point cloud of acquisition.
In general, after obtaining a cloud, periphery can all generate some rough error points.To the method class of these rough error points removal
Like k nearest neighbor.Assuming that giving Radius R centered on the certain point in cloud, the point number n with its neighbour being found out in R, if n
Less than some threshold value, then it can determine that the point is rough error point.But it due to a cloud enormous amount, is all needed before handling it
Establish spatial index.Common space index structure has a KD tree, R tree, quaternary tree, Octree etc., and in these structures, KD
Tree is most widely used with Octree, and the Octree module in the library Selection utilization PCL here, and rough error point is carried out to it and is gone
It removes.Class pcl::octree::OctreePointCloud is the octree structure realized for the big problem of point cloud data amount, should
Class derives from many subclasses, to facilitate to the existing different processing of mysorethorn.Class pcl::octree::
OctreePointCloudSearch realizes point cloud neighbour's effective search algorithm based on Octree, the member used herein
Function is radiusSearch (), mainly neighbour's point set number in acquisition radius radius, by with the threshold that is set in advance
Value size is compared, and then determines whether the point is rough error point, once sentencing as rough error point, is just deleted in the Octree
The point carries out filtering out work.The parameter setting of processing are as follows: the resolution ratio of octree structure is 0.000008f, and radius is set as
0.01f, neighbor point number threshold value are set as 13.
The invention has the benefit that method of the invention carries out face three-dimensional point using existing Kinect depth camera
The optimization process flow of cloud avoids the expensive price of traditional high precision apparatus or the problem that general algorithm precision is not high；
Inside the depth optimization the step of, when carrying out surface recovery using depth after obtained optimization, general utilization method is not taken
Vector is directly merged, because the fusion of normal vector solves surface and can generate offset error, here with concentration gradient and method
The geometrical relationship of vector is solved, and last depth is directly obtained, and finally obtains quality higher three using the variation of coordinate
Dimension point cloud.
Detailed description of the invention
Fig. 1 is that the present invention is based on the flow diagrams of the face three-dimensional point cloud optimized treatment method of kinect camera；
Fig. 2 is depth optimization Processing Algorithm flow chart.
Specific embodiment
Summary is described in detail technical solution of the present invention, and details are not described herein.
Claims (2)
1. a kind of face three-dimensional point cloud optimized treatment method based on kinect camera, which comprises the following steps:
S1, kinect camera calibration, the inside and outside parameter including depth camera and color camera；
S2, tested facial image: the camera in applying step S1 is obtained, obtains the RGB figure and depth D figure of tested face, and will
The RGB figure and depth D figure of acquisition carry out registration process, i.e., are aligned the different depth D figure of pixel size and RGB figure one by one, make
The depth value of each pixel has a corresponding rgb value；
S3, the image of acquisition is pre-processed, the missing point including depth D figure, the erroneous point of RGB figure, and to depth D figure into
The processing of row bilateral filtering；Method particularly includes:
For the missing point of depth D figure: assuming that missing pixel point coordinate is (i, j), from the four direction up and down of the point into
Then row search averages four values until finding the first nonzero value in each direction to fill lacking for the pixel
Mistake value；
For the erroneous point of RGB figure: judging the value in three channels of RGB figure respectively first, if three channels are simultaneously 0, decide that
The point is erroneous point, and then assigning three channels of the pixel is all 1；
S4, face is extracted from the RGB figure of acquisition, and the face extracted is projected to the depth D figure the inside of alignment, obtain
Corresponding human face region in depth D figure；Method particularly includes:
S41, the RGB figure of acquisition is mapped:
RGB color is mapped to YCbCr space by following formula 1:
S42, face is tentatively extracted:
Use in the transformation space image that dimensional Gaussian model inspection S41 is obtained shown in following formula 2 each pixel for
The probability of the colour of skin:
After the probability value for obtaining each pixel, binaryzation is carried out to the probability graph of acquisition, is as a result preliminary extraction face
As a result；
S43, the bianry image of acquisition is handled, extracts target face rectangular area；
S5, RGB and depth D are combined, depth optimization processing is carried out to human face region corresponding in D figure, recycles depth
Depth after optimization is converted into three-dimensional point cloud by the internal reference of camera, method particularly includes:
S51, eigen decomposition is carried out to RGB image, is decomposed according to illumination model shown in following formula 3:
In formula 3,It is the intensity at image slices vegetarian refreshments (i, j), that is, grey scale pixel value,It is pixel
Light and shade value, ρ (i, j) indicate multiple reflection rate, and β (i, j) indicates local illumination variation impact factor；
The parameter obtained in S52, solution procedure S51, specifically:
S521, light and shade value is solved
The light and shade value of image certain point is expressed as to the linear forms of the normal vector of the point shown in formula 4:
Wherein,It is the normal vector of each pixel of image,It is first four spheric harmonic function coefficient,In image
Each pixel may serve to solve light and shade value, therefore, need exist for solve be overdetermination least square coefficient estimation ask
Topic, following formula 5:
S522, multiple reflection rate ρ (i, j) is solved:
It is solved using energy function shown in following formula 6:
In formula 6, N is the neighborhood of certain pixel,It is weight coefficient relevant to pixel intensity value,It is and initial depth phase
The weight coefficient of pass, I are gray values, and subscript k is pixel number, are expressed as formula 7:
In formula 7, z (i, j) is the depth value of pixel (i, j), and I (i, j) is the rgb value of pixel (i, j), σcIt is about certain picture
The coefficient of element and pixel intensity value in its neighborhood, σdIt is about the discontinuous related coefficient of depth；
S523, part illumination variation impact factor β (i, j) is solved:
β (i, j) is obtained by following formula 8:
S53, the depth optimization that face is carried out according to the parameter that step S52 is obtained, are minimized using energy, pass through following formula 9
It solves:
In formula 9, z0It is the depth value that image initial obtains, Δ indicates its Laplace operator；△ is the local derviation relationship of depth,
It is the expression for the normal vector of each pixel, there is relationship shown in following formula 10:
In formula 10, the subscript x and y of z respectively indicates the gradient in the direction x and the gradient in the direction y；
The method for solving of formula 9 specifically:
S531, fixed nonlinear terms, i.e. denominator term when solving normal vector, this denominator term is tied with previous calculating
Fruit brings normal vector this time into as given value and calculates, shown in following formula 11:
S532, using obtained normal vector, update linear eigen decomposition illumination model item, following formula 12:
S533, iterative solution, according to parameter z0,ρ, β, as f (zk-1)>f(zk) set up when, always iterate to calculate:
Until f (zk+1)>f(zk), stop iterative process, previous zkThat is the obtained final result of optimization processing；
Described τ, σc、σd、λρ、For preset weight threshold；
S54, according to the parameter of Kinect depth camera, using the conversion of coordinate system by the obtained two-dimensional depth z of S533 iterationk
It is transformed into three-dimensional point cloud form, following formula 13:
In formula 13, (i, j) is the pixel in image, [x(i,j),y(i,j),z(i,j)] it is the three-dimensional seat of pixel (i, j) to be solved
Mark, zk(i, j) is its depth value, K-1It is inverting for depth camera internal reference matrix.
2. a kind of face three-dimensional point cloud optimized treatment method based on kinect camera according to claim 1, feature
It is, further includes:
S6, the processing of rough error point is removed to the three-dimensional point cloud of acquisition:
Using the Octree module in the library PCL, the removal of rough error point is carried out to the three-dimensional point cloud of acquisition.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710464550.2A CN107169475B (en) | 2017-06-19 | 2017-06-19 | A kind of face three-dimensional point cloud optimized treatment method based on kinect camera |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710464550.2A CN107169475B (en) | 2017-06-19 | 2017-06-19 | A kind of face three-dimensional point cloud optimized treatment method based on kinect camera |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107169475A CN107169475A (en) | 2017-09-15 |
| CN107169475B true CN107169475B (en) | 2019-11-19 |
Family
ID=59819441
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710464550.2A Expired - Fee Related CN107169475B (en) | 2017-06-19 | 2017-06-19 | A kind of face three-dimensional point cloud optimized treatment method based on kinect camera |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107169475B (en) |
Families Citing this family (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109948400A (en) * | 2017-12-20 | 2019-06-28 | 宁波盈芯信息科技有限公司 | It is a kind of to be able to carry out the smart phone and its recognition methods that face characteristic 3D is identified |
| CN108564041B (en) * | 2018-04-17 | 2020-07-24 | 云从科技集团股份有限公司 | Face detection and restoration method based on RGBD camera |
| CN109087347B (en) * | 2018-08-15 | 2021-08-17 | 浙江光珀智能科技有限公司 | Image processing method and device |
| CN109194943B (en) * | 2018-08-29 | 2020-06-02 | 维沃移动通信有限公司 | Image processing method and terminal equipment |
| CN109583304A (en) * | 2018-10-23 | 2019-04-05 | 宁波盈芯信息科技有限公司 | A kind of quick 3D face point cloud generation method and device based on structure optical mode group |
| CN111344741B (en) * | 2019-01-31 | 2023-06-20 | 深圳市瑞立视多媒体科技有限公司 | Data missing processing method and device for three-dimensional track data |
| CN110059537A (en) * | 2019-02-27 | 2019-07-26 | 视缘(上海)智能科技有限公司 | A kind of three-dimensional face data acquisition methods and device based on Kinect sensor |
| CN111696145B (en) * | 2019-03-11 | 2023-11-03 | 北京地平线机器人技术研发有限公司 | Depth information determining method, depth information determining device and electronic equipment |
| CN109816786B (en) * | 2019-03-28 | 2022-11-25 | 深圳市超准视觉科技有限公司 | Three-dimensional point cloud reconstruction method and device and computer equipment |
| CN110009676B (en) * | 2019-04-11 | 2019-12-17 | 电子科技大学 | Intrinsic property decomposition method of binocular image |
| CN110288581B (en) * | 2019-06-26 | 2022-11-04 | 电子科技大学 | Segmentation method based on model for keeping shape convexity level set |
| CN110276316B (en) * | 2019-06-26 | 2022-05-24 | 电子科技大学 | Human body key point detection method based on deep learning |
| CN112562082A (en) * | 2020-08-06 | 2021-03-26 | 长春理工大学 | Three-dimensional face reconstruction method and system |
| CN114913287B (en) * | 2022-04-07 | 2023-08-22 | 北京拙河科技有限公司 | Three-dimensional human body model reconstruction method and system |
| CN115953400B (en) * | 2023-03-13 | 2023-06-02 | 安格利(成都)仪器设备有限公司 | Corrosion pit automatic detection method based on three-dimensional point cloud object surface |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103353935A (en) * | 2013-07-19 | 2013-10-16 | 电子科技大学 | 3D dynamic gesture identification method for intelligent home system |
| CN104463880A (en) * | 2014-12-12 | 2015-03-25 | 中国科学院自动化研究所 | RGB-D image acquisition method |
| CN105469042A (en) * | 2015-11-20 | 2016-04-06 | 天津汉光祥云信息科技有限公司 | Improved face image comparison method |
| WO2016107638A1 (en) * | 2014-12-29 | 2016-07-07 | Keylemon Sa | An image face processing method and apparatus |
| CN106780592A (en) * | 2016-06-30 | 2017-05-31 | 华南理工大学 | Kinect depth reconstruction algorithms based on camera motion and image light and shade |
-
2017
- 2017-06-19 CN CN201710464550.2A patent/CN107169475B/en not_active Expired - Fee Related
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103353935A (en) * | 2013-07-19 | 2013-10-16 | 电子科技大学 | 3D dynamic gesture identification method for intelligent home system |
| CN104463880A (en) * | 2014-12-12 | 2015-03-25 | 中国科学院自动化研究所 | RGB-D image acquisition method |
| WO2016107638A1 (en) * | 2014-12-29 | 2016-07-07 | Keylemon Sa | An image face processing method and apparatus |
| CN105469042A (en) * | 2015-11-20 | 2016-04-06 | 天津汉光祥云信息科技有限公司 | Improved face image comparison method |
| CN106780592A (en) * | 2016-06-30 | 2017-05-31 | 华南理工大学 | Kinect depth reconstruction algorithms based on camera motion and image light and shade |
Non-Patent Citations (1)
| Title |
|---|
| 基于RGBD的人体行为识别系统;周康;《中国优秀硕士学位论文全文数据库 (信息科技辑)》;20160215(第2期);第13-14、45-47页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107169475A (en) | 2017-09-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107169475B (en) | A kind of face three-dimensional point cloud optimized treatment method based on kinect camera | |
| CN108549873B (en) | Three-dimensional face recognition method and three-dimensional face recognition system | |
| CN106803267B (en) | Kinect-based indoor scene three-dimensional reconstruction method | |
| CN104376552B (en) | A kind of virtual combat method of 3D models and two dimensional image | |
| CN101443817B (en) | Method and device for determining correspondence, preferably for the three-dimensional reconstruction of a scene | |
| GB2581374A (en) | 3D Face reconstruction system and method | |
| CN104335005B (en) | 3D is scanned and alignment system | |
| CN101996407B (en) | Colour calibration method for multiple cameras | |
| JP2009020761A (en) | Image processing apparatus and method thereof | |
| CN106157372A (en) | A kind of 3D face grid reconstruction method based on video image | |
| CN111523398A (en) | Method and device for fusing 2D face detection and 3D face recognition | |
| CN103914699A (en) | Automatic lip gloss image enhancement method based on color space | |
| CN109754459B (en) | Method and system for constructing human body three-dimensional model | |
| WO2010041584A1 (en) | Imaging system and method | |
| CN110263768A (en) | A kind of face identification method based on depth residual error network | |
| CN106705849A (en) | Calibration method of linear-structure optical sensor | |
| CN110567441B (en) | Particle filter-based positioning method, positioning device, mapping and positioning method | |
| CN111144213A (en) | Object detection method and related equipment | |
| CN101976436A (en) | Pixel-level multi-focus image fusion method based on correction of differential image | |
| CN107292269A (en) | Facial image false distinguishing method, storage, processing equipment based on perspective distortion characteristic | |
| Fang et al. | Laser stripe image denoising using convolutional autoencoder | |
| CN107610219A (en) | The thick densification method of Pixel-level point cloud that geometry clue perceives in a kind of three-dimensional scenic reconstruct | |
| CN112330813A (en) | Wearing three-dimensional human body model reconstruction method based on monocular depth camera | |
| CN115880415A (en) | Three-dimensional reconstruction method and device, electronic equipment and storage medium | |
| Condorelli et al. | A comparison between 3D reconstruction using nerf neural networks and mvs algorithms on cultural heritage images |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191119 Termination date: 20200619 |