CN104935832B - For the video keying method with depth information - Google Patents
For the video keying method with depth information Download PDFInfo
- Publication number
- CN104935832B CN104935832B CN201510151211.XA CN201510151211A CN104935832B CN 104935832 B CN104935832 B CN 104935832B CN 201510151211 A CN201510151211 A CN 201510151211A CN 104935832 B CN104935832 B CN 104935832B
- Authority
- CN
- China
- Prior art keywords
- value
- frame
- pixel
- video
- prospect
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses the video keying methods with depth information of being directed to, three components including calculating every frame image;Video is segmented, the interframe in each video segmentation is made to keep coherent;Obtain the prospect and transparency estimated value that each pixel is optimal in the zone of ignorance of each frame;Obtain the global optimization solution of pixel in all zone of ignorances in the video segmentation;Complete the stingy as processing of whole section of video.The foreground target that the present invention is suitable for carrying out the video sequence with interframe continuity rapidly and efficiently extracts, and is able to maintain the space-time consistency of video keying, reduces flashing and vision mutation, improves and scratch as computational efficiency.
Description
[technical field]
The invention belongs to field of image processings, especially for the video keying method for having depth information.
[background technique]
Taking for image has many multiduty technologies always by scholar and enterprise as one in computer vision field
The concern of industry has a large amount of successful practical applications in many fields.And interesting target is taken from the image sequence of video
Technology be the deeper time research of picture is scratched to single image, though the starting stage is at present, because of its broad application prospect
So that more and more scholars start the research for being dedicated to the field.
Single-frame images scratch as process be calculate image in each pixel foreground pixel value F, background pixel value B and
Alpha transparence value is underconstrained problem, is calculated complicated.And video keying is the stingy picture to image sequence, involved problem is more
Complexity, there are the difficult points of several aspects to need to solve, and emphasis has: 1. data volumes to be processed are huge, need efficient process video
A large amount of pixels in sequence are improved and are scratched as efficiency;2. video keying needs to keep the space-time consistency between sequence, reduces and scratch picture
Generate flashing and vision mutation;
Currently, existing video keying method mainly divides following a few classes:
Stingy picture algorithm frame by frame.Image sequence in video is regarded as independent picture frame by this algorithm, then to each
Frame image takes algorithm using existing single image to realize and scratch as process.This method is more convenient, easy to accomplish, can be by
Lead to the correlation for not considering adjacent interframe properly in the independent processing to every frame image, can make between continuous sequence image
Corresponding pixel transparent angle value generates difference, is unable to ensure the interframe continuity for taking result, generates flashing and vision mutation.
The stingy image space method of 3 D stereo.Video sequence is regarded as a three-dimensional solid by such algorithm, by three-dimensional geometry
One entirety of stereoscopic work is handled scratch picture.This method generally requires progress multistep and scratches as ideal to obtain one
As a result, it is inaccurate to scratch the effect obtained as result since it is preliminary, the optimization processing for needing to continue obtains better effect,
Therefore it scratches as efficiency is lower.
The stingy picture algorithm of successive frame.Such method equally applies some more mature single width and scratches image space method, simultaneously will
Between consecutive frame scratch as result as the restrictive condition of interframe continuity be applied to present frame it is stingy picture step in.Due to considering
The association of the adjacent interframe in front and back can obtain relatively good stingy as effect.But due to all using single frames to scratch picture all frames,
Time spent by such method is more.
The present invention is based on the interframe continuities using video, design estimation and the optimization side of a kind of quick prospect and transparency
Method extracts the foreground target interested in video sequence.Automatic three based on depth information have been used in the method
Component generates and video segmentation, quick prospect and transparence value estimation, and has carried out optimization and the optimization of interframe boundling in frame, most
Ideal scratch as effect is obtained eventually.
[summary of the invention]
It is an object of the invention to solve the problems, such as the above-mentioned of the existing field, a kind of sequence of video images is provided and is quickly scratched
Image space method, the foreground target for being suitable for carrying out the video sequence with interframe continuity rapidly and efficiently extracts, including following step
It is rapid:
S01 calculates three components of every frame image;
S02 is segmented video, and the interframe in each video segmentation is made to keep coherent,
Step S02 includes: the area ratio that is overlapped between the zone of ignorance calculated between the successive frame of front and backChoose PU< 0.8 frame is key frame Frkey;By certain key frame Frkey,i
And key frame Frkey,iWith next key frame Frkey,i+1Between image sequence be grouped into same segmentation Seg [Frkey,i]=
[Frkey,i,Fr0,Fr1,....Frn,Frkey,i+1);Wherein Frt, the normal frames of t=0,1,2..., n between key frame;
S03 obtains the prospect that each pixel is optimal in the zone of ignorance of each frame and transparency estimated value, specifically, by following
Mode is realized: scratching picture to the two frame application closed type of first and last of video segmentation;Kinematic parameter based on optical flow method principle solving pixel;
Estimate the prospect value and transparence value of pixel in remaining each frame zone of ignorance frame by frame in conjunction with interframe continuity;It is calculated with simulated annealing
Method optimizes the prospect value and transparence value estimated in remaining each frame;
S04 obtains the global optimization solution of pixel prospect value and transparence value in all zone of ignorances in the video segmentation, leads to
It crosses and solves energy equation minimum using gradient descent method to carry out:
Energy equation are as follows:
Wherein FrmFor the picture frame in step 3, Δ αk t1→t2With Δ Fk t1→t2Respectively indicate FrmZone of ignoranceIn
The prospect of pixel undetermined and transparence value pushing forward or backward frame difference during leading, With
For FrmIn k-th pixel undetermined by pushing forward or backward the prospect value for leading to respectively obtain and transparence value;For the coefficient for meeting normal distribution, in FrmPlace reaches peak value, and N is control constant;
S05 is according to the global optimums of all video segmentations (F ', α ') value;Complete the stingy as processing of whole section of video.
Further, three components described in step S01, which refer to, divides an image into three parts region, determining foreground area RF、
Determining background area RBAnd zone of ignorance R between the twoU, this method be based on depth information calculate three components automatically:
1) depth threshold is selected, depth map is split and obtains a binarization segmentation as a result, being less than threshold value
Region is as prospect;
2) morphological erosion operation is carried out to the foreground area of binaryzation, obtained region is as determining foreground zone after corrosion
Domain RF;
3) morphological dilation is carried out to binaryzation foreground area and negated, gained region is as determining background area
RB;
It 4) is zone of ignorance R between determining prospect, background areaU, calculate its area SU。
Preferably, the two frame application closed type of first and last to video segmentation scratches picture, it is realized by the following method: right
Seg[Frkey,i] in key frame Frkey,i, image space method is scratched based on classical closed type and obtains its zone of ignorance RUIn each pixel
Prospect, background and transparency (F, B, α) value.
Preferably, the kinematic parameter based on optical flow method principle solving pixel, is realized by the following method:
Enabling I is the gray value of image, and D is the depth value of image, according to the basic principle of optical flow method, to each pixel
It is all satisfied following two formula:
Ix·u+Iy·v+It=0 (1)
Dx·u+Dy·v+Dt=0 (2)
Wherein Ix, Iy, ItAnd Dx, Dy, DtRespectively partial derivative of the sum of the grayscale values depth value on x, the direction y and time t,
It can be directly calculated by the color diagram and depth map of frame sequence;
For velocity component of the pixel on x, the direction y;It, can by solving above-mentioned two formula
Acquire the kinematic parameter of each pixelWhereinRepresentation speed size,
Representation speed direction.
Preferably, the combination interframe continuity estimates the information of pixel in each frame zone of ignorance frame by frame, by with
Lower method is realized:
For present frame FrtZone of ignoranceIn k-th of pixelAccording to its kinematic parameter meterIt is calculated in Frt-1In positionIt usesThe background value at placeAsBackground estimating valueIt usesLocate prospect, background and the transparency of each pixel in 3 × 3 neighborhoods
To estimateThe prospect value at placeAnd transparence value
Preferably, described optimize the parameter estimated in each frame with simulated annealing, by with lower section
Method is realized:
It is assumed that pk tPixel background color value at (x, y)It is constant, using simulated annealing to foreground pixel valueWith
Transparence valueIt optimizes,
Wherein, the solution that simulated annealing optimization algorithm uses is S are as follows:
Wherein, Δ σα=σα/ (3N), Δ σF=σF/(3·N);σFAnd σαRespectively pk t-13 × 3 neighborhoods at (x ', y ')
The variance of interior prospect and transparence value;N is constant, for controlling step-length;
The evaluation function C (S) that simulated annealing optimization algorithm uses are as follows:
Wherein, β1, β2, β3For constant factor, Color (x, y) is the color RGB vector of pixel,WithFor
The estimated value of initial background, prospect and transparency,WithThe prospect and transparence value found out for current iteration,
Current iteration includes:
It is searched for from solution space new solution (α *, F*), (α *, F*) ∈ S;
Incremental computations
If Δ t ' < 0, receive (α ', F ') as current new explanation, otherwise received with probability exp (- Δ t '/T) (α *,
F*) as current new explanation;
Currently solution is optimal solution if meeting termination condition, terminator, termination condition be several continuous new explanations all not
Received situation.
The invention has the following advantages: being suitable for carrying out rapidly and efficiently the video sequence with interframe continuity
Foreground target extracts, and is able to maintain the space-time consistency of video keying, reduces flashing and vision mutation, improves and scratch as calculating effect
Rate.
[Detailed description of the invention]
The present invention will be further explained below with reference to the attached drawings:
Fig. 1 is the flow chart for the video keying method with depth information;
The motion change figure of three components of Fig. 2 video segmentation frame and zone of ignorance pixel in interframe.
[specific embodiment]
Combined with specific embodiments below, and in conjunction with attached drawing, further description of the technical solution of the present invention:
Combined with specific embodiments below, and in conjunction with attached drawing, further description of the technical solution of the present invention:
Embodiment 1: video is shot using depth camera (Kinect of such as Microsoft), obtains the image sequence with depth information
Column, i.e. image also have depth Depth value other than color Color value.
Parsing obtains image sequence, applies stingy image space method of the invention to the image sequence, carries out in accordance with the following steps:
(1) in sequence of computed images every frame image three components;
(2) key frame extraction and video segmentation;According to the area ratio P being overlapped between the zone of ignorance between consecutive frameU<0.8
Frame be key frame Frkey, by certain key frame Frkey,iWith next key frame Frkey,i+1Between image sequence be grouped into same segmentation
Seg[Frkey,i];
(3) pixel prospect, background and the transparency (F, B, α) in video segmentation in every frame zone of ignorance RU are calculated frame by frame
Value.The simulated annealing optimization algorithm flow used during calculating are as follows:
1. initialization: initial temperature T, initial value C (x, y), the number of iterations L of each T value;
2. couple k=1 ..., L executes the 3rd to step 6:
3. new solution (α ', F ') is searched for from solution space, (α ', F ') ∈ S;
4. incremental computations
5. if Δ t ' < 0, receives (α ', F ') as current new explanation, otherwise with probability exp (- Δ t '/T) receiving
(α ', F ') as current new explanation.
6. currently solution is optimal solution, terminator if meeting termination condition.Termination condition be several continuous new explanations all
Unaccredited situation.
7.T is reduced, and T is intended to 0, turns step 2.
(4) interframe boundling optimizes.Boundling optimization process solves following equation by using gradient descent method to carry out:
(5) according to the global optimum of all video segmentations (F ', α ') value, in addition the background value being held essentially constant, is completed
The stingy picture processing of whole section of video.
Compared with the background art, innovation of the invention is:
1) video segmentation.Calculate three components automatically using depth information, the degree of overlapping based on consecutive frame zone of ignorance is to view
Frequency is segmented, and the interframe continuity in video segmentation is utmostly maintained;
2) single frames is scratched picture+interframe and is calculated.In video segmentation, it is only necessary to two frame of first and last is scratched using single frames as technology,
The mode that remaining frame must be estimated and be optimized based on kinematic parameter progress prospect and transparence value carries out scratching picture, improves to scratch and imitate as calculating
Rate.
3) double optimization process.In individual picture frame, it is utilized the thought of simulated annealing, carries out prospect color value and thoroughly
The optimal screening of bright value;Boundling optimization, foreground and transparence value are carried out when interframe is kept in the sequence interframe of video segmentation
Empty consistency, so that final is stingy as result is more robust reliable.
Above-mentioned specific embodiment is used to illustrate the present invention, rather than limits the invention, of the invention
In spirit and scope of protection of the claims, to any modifications and changes that the present invention makes, protection model of the invention is both fallen within
It encloses.
Claims (6)
1. the video keying method with depth information of being directed to, which comprises the following steps:
S01 calculates three components of every frame image;
S02 is segmented video, and the interframe in each video segmentation is made to keep coherent,
Step S02 includes: the area ratio that is overlapped between the zone of ignorance calculated between the successive frame of front and backChoose PU< 0.8 frame is key frame Frkey;By certain key frame Frkey,i
And key frame Frkey,iWith next key frame Frkey,i+1Between image sequence be grouped into same segmentation Seg [Frkey,i]=
[Frkey,i,Fr0,Fr1,....Frn,Frkey,i+1);Wherein Frt, the normal frames of t=0,1,2..., n between key frame;
S03 obtains the prospect that each pixel is optimal in the zone of ignorance of each frame and transparency estimated value, specifically, in the following manner
It realizes: picture is scratched to the two frame application closed type of first and last of video segmentation;Kinematic parameter based on optical flow method principle solving pixel;In conjunction with
Interframe continuity estimates the prospect value and transparence value of pixel in remaining each frame zone of ignorance frame by frame;With simulated annealing pair
Prospect value and the transparence value optimization that estimation obtains in remaining each frame;
S04 obtains the global optimization solution of pixel prospect value and transparence value in all zone of ignorances in the video segmentation, by making
Energy equation minimum is solved with gradient descent method to carry out:
Energy equation are as follows:
Wherein FrmFor the picture frame in step 3, Δ αk t1→t2With Δ Fk t1→t2Respectively indicate FrmZone of ignorance RU mIn picture undetermined
The prospect and transparence value of element are pushing forward or backward frame difference during leading, Fk m、Fk m’、αk mAnd αk m’For FrmIn k-th to
Determine pixel and leads the prospect value respectively obtained and transparence value by pushing forward or backward;To meet normal state
The coefficient of distribution, in FrmPlace reaches peak value, and N is control constant;
S05 completes the stingy as processing of whole section of video according to the global optimums of all video segmentations (F ', α ') value.
2. according to claim 1 be directed to the video keying method with depth information, it is characterised in that: described in step S01
Three components, which refer to, divides an image into three parts region, determining foreground area RF, determine background area RBAnd between the two
Zone of ignorance RU, this method be based on depth information calculate three components automatically:
1) depth threshold is selected, depth map is split and obtains a binarization segmentation as a result, being less than the region of threshold value
As prospect;
2) morphological erosion operation is carried out to the foreground area of binaryzation, obtained region is as determining foreground area R after corrosionF;
3) morphological dilation is carried out to binaryzation foreground area and negated, gained region is as determining background area RB;
It 4) is zone of ignorance R between determining prospect, background areaU, calculate its area SU。
3. according to claim 1 be directed to the video keying method with depth information, which is characterized in that described to video point
The two frame application closed type of first and last of section scratches picture, is realized by the following method: to Seg [Frkey,i] in key frame Frkey,i, it is based on
Classical closed type scratches image space method and obtains its zone of ignorance RUIn each pixel prospect, background and transparency (F, B, α) value.
4. according to claim 1 be directed to the video keying method with depth information, which is characterized in that described to be based on light stream
The kinematic parameter of method principle solving pixel, is realized by the following method:
Enabling I is the gray value of image, and D is the depth value of image, full to each pixel according to the basic principle of optical flow method
Sufficient following two formula:
Ix·u+Iy·v+It=0 (1)
Dx·u+Dy·v+Dt=0 (2)
Wherein Ix, Iy, ItAnd Dx, Dy, DtRespectively partial derivative of the sum of the grayscale values depth value on x, the direction y and time t, can be with
It is directly calculated by the color diagram and depth map of frame sequence;
For velocity component of the pixel on x, the direction y;By solving above-mentioned two formula, can acquire
The kinematic parameter of each pixelWhereinRepresentation speed size,It represents
Directional velocity.
5. according to claim 3 be directed to the video keying method with depth information, which is characterized in that the combination interframe
Continuity estimates the prospect value and transparence value of pixel in remaining each frame zone of ignorance frame by frame, is realized by the following method:
For present frame FrtZone of ignorance RU tIn k-th of pixel pk t(x, y), according to its kinematic parameterIt calculates
It is in Frt-1In position pk t-1(x ', y '), use pk t-1Background value B at (x, y)k t-1(x, y) is used as Pk tThe background of (x, y)
Estimated valueUse pk t-1Prospect, background and the transparency of each pixel estimate p in 3 × 3 neighborhoods at (x ', y ')k t(x,y)
The prospect value at placeAnd transparence value
6. according to claim 1 be directed to the video keying method with depth information, which is characterized in that described to be moved back with simulation
Fiery algorithm optimizes the prospect value and transparence value estimated in remaining each frame, is realized by the following method:
It is assumed that pk tPixel background color value at (x, y)It is constant, using simulated annealing to foreground pixel valueWith it is transparent
Angle valueIt optimizes,
Wherein, the solution that simulated annealing optimization algorithm uses is S are as follows:
Wherein, Δ σα=σα/ (3N), Δ σF=σF/(3·N);σFAnd σαRespectively pk t-1Prospect in 3 × 3 neighborhoods at (x ', y ')
With the variance of transparence value;N is constant, for controlling step-length;
The evaluation function C (S) that simulated annealing optimization algorithm uses are as follows:
Wherein, β1, β2, β3For constant factor, Color (x, y) is the color RGB vector of pixel,WithIt is initial
The estimated value of background, prospect and transparency,WithThe prospect and transparence value found out for current iteration,
Current iteration includes:
It is searched for from solution space new solution (α *, F*), (α *, F*) ∈ S;
Calculate increment Delta t '=Cost (α *, F*)-C (x, y), Cost (α *, F*)=α * F*+ (1- α *)
If Δ t ' < 0, receives (α *, F*) as current new explanation, (α *, F*) is otherwise received with probability exp (- Δ t '/T) and is made
For current new explanation;
Currently solution is optimal solution if meeting termination condition, and terminator, termination condition is that several continuous new explanations are not all connect
By the case where.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510151211.XA CN104935832B (en) | 2015-03-31 | 2015-03-31 | For the video keying method with depth information |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201510151211.XA CN104935832B (en) | 2015-03-31 | 2015-03-31 | For the video keying method with depth information |
Publications (2)
Publication Number | Publication Date |
---|---|
CN104935832A CN104935832A (en) | 2015-09-23 |
CN104935832B true CN104935832B (en) | 2019-07-12 |
Family
ID=54122773
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201510151211.XA Active CN104935832B (en) | 2015-03-31 | 2015-03-31 | For the video keying method with depth information |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN104935832B (en) |
Families Citing this family (14)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN106204567B (en) * | 2016-07-05 | 2019-01-29 | 华南理工大学 | A kind of natural background video matting method |
CN106331533A (en) * | 2016-08-10 | 2017-01-11 | 深圳市企拍文化科技有限公司 | Method for adding LOGO in video |
CN107018322B (en) * | 2017-03-09 | 2020-02-11 | Oppo广东移动通信有限公司 | Control method and control device for rotary camera auxiliary composition and electronic device |
CN106993112B (en) * | 2017-03-09 | 2020-01-10 | Oppo广东移动通信有限公司 | Background blurring method and device based on depth of field and electronic device |
CN107133964B (en) * | 2017-06-01 | 2020-04-24 | 江苏火米互动科技有限公司 | Image matting method based on Kinect |
CN107481261B (en) * | 2017-07-31 | 2020-06-16 | 中国科学院长春光学精密机械与物理研究所 | Color video matting method based on depth foreground tracking |
CN108154086B (en) * | 2017-12-06 | 2022-06-03 | 北京奇艺世纪科技有限公司 | Image extraction method and device and electronic equipment |
CN109903291B (en) * | 2017-12-11 | 2021-06-01 | 腾讯科技(深圳)有限公司 | Image processing method and related device |
CN111882576A (en) * | 2018-04-17 | 2020-11-03 | 芜湖岭上信息科技有限公司 | Method and device for classifying depth information of foreground pixels of video image and segmenting foreground |
CN113766319A (en) * | 2018-06-01 | 2021-12-07 | 北京市商汤科技开发有限公司 | Image information processing method and device, and storage medium |
WO2020062898A1 (en) * | 2018-09-26 | 2020-04-02 | 惠州学院 | Video foreground target extraction method and apparatus |
CN110111342B (en) * | 2019-04-30 | 2021-06-29 | 贵州民族大学 | Optimized selection method and device for matting algorithm |
CN113194270B (en) * | 2021-04-28 | 2022-08-05 | 北京达佳互联信息技术有限公司 | Video processing method and device, electronic equipment and storage medium |
CN113610865B (en) * | 2021-07-27 | 2024-03-29 | Oppo广东移动通信有限公司 | Image processing method, device, electronic equipment and computer readable storage medium |
Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101673400A (en) * | 2008-09-08 | 2010-03-17 | 索尼株式会社 | Image processing apparatus, method, and program |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8897562B2 (en) * | 2012-06-29 | 2014-11-25 | Adobe Systems Incorporated | Adaptive trimap propagation for video matting |
-
2015
- 2015-03-31 CN CN201510151211.XA patent/CN104935832B/en active Active
Patent Citations (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101673400A (en) * | 2008-09-08 | 2010-03-17 | 索尼株式会社 | Image processing apparatus, method, and program |
Non-Patent Citations (4)
Title |
---|
一种鲁棒视频抠图算法;李闻;《计算机应用研究》;20100131;第27卷(第1期);第358-376页 * |
基于Kinect的抠像算法研究;张约伦;《中国优秀硕士学位论文全文数据库》;20140131;第I138-2123页 * |
改进的自然图像鲁棒抠图算法;黄睿,王翔;《计算机工程与应用》;20131231;第49卷(第12期);第136-139页 * |
视频抠图算法的研究;彭浩浩;《中国优秀硕士学位论文全文数据库》;20131031;第I138-411页 * |
Also Published As
Publication number | Publication date |
---|---|
CN104935832A (en) | 2015-09-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN104935832B (en) | For the video keying method with depth information | |
US11017586B2 (en) | 3D motion effect from a 2D image | |
CN107578436B (en) | Monocular image depth estimation method based on full convolution neural network FCN | |
Stühmer et al. | Real-time dense geometry from a handheld camera | |
US8610712B2 (en) | Object selection in stereo image pairs | |
US8447098B1 (en) | Model-based stereo matching | |
US8619073B2 (en) | System and method for recovering three-dimensional particle systems from two-dimensional images | |
US9076258B2 (en) | Stylizing animation by example | |
CN107481261A (en) | A kind of color video based on the tracking of depth prospect scratches drawing method | |
CN101923719B (en) | Particle filter and light stream vector-based video target tracking method | |
CN110189339A (en) | The active profile of depth map auxiliary scratches drawing method and system | |
WO2013178725A1 (en) | Segmentation of a foreground object in a 3d scene | |
US20180005039A1 (en) | Method and apparatus for generating an initial superpixel label map for an image | |
US20180247418A1 (en) | Method and apparatus for object tracking and segmentation via background tracking | |
CN105023246B (en) | A kind of image enchancing method based on contrast and structural similarity | |
CN105590327A (en) | Motion estimation method and apparatus | |
Chang et al. | Topology-constrained layered tracking with latent flow | |
CN104159098B (en) | The translucent edge extracting method of time domain consistence of a kind of video | |
Woodford et al. | On New View Synthesis Using Multiview Stereo. | |
Xu et al. | Video-object segmentation and 3D-trajectory estimation for monocular video sequences | |
Zhang et al. | Dehazing with improved heterogeneous atmosphere light estimation and a nonlinear color attenuation prior model | |
Paz et al. | A variational approach to online road and path segmentation with monocular vision | |
CN112561995B (en) | Real-time and efficient 6D attitude estimation network, construction method and estimation method | |
Zhang et al. | High-quality stereo video matching via user interaction and space-time propagation | |
Yan et al. | Re-texturing by intrinsic video |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |