CN110047091A - One kind is based on the estimation of camera track and the matched digital image stabilization method of characteristic block - Google Patents
One kind is based on the estimation of camera track and the matched digital image stabilization method of characteristic block Download PDFInfo
- Publication number
- CN110047091A CN110047091A CN201910192526.7A CN201910192526A CN110047091A CN 110047091 A CN110047091 A CN 110047091A CN 201910192526 A CN201910192526 A CN 201910192526A CN 110047091 A CN110047091 A CN 110047091A
- Authority
- CN
- China
- Prior art keywords
- frame
- estimation
- camera
- video
- transformation matrix
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/207—Analysis of motion for motion estimation over a hierarchy of resolutions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/246—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments
- G06T7/251—Analysis of motion using feature-based methods, e.g. the tracking of corners or segments involving models
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10016—Video; Image sequence
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Studio Devices (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses one kind based on the estimation of camera track and the matched digital image stabilization method of characteristic block, characteristic point is extracted using SIFT algorithm and is slightly matched using characteristic point, using use M estimation sampling consistency (M-estimator Sample and Consensus, MSAC) algorithm remove Exceptional point.By obtained matching to being fitted two-dimensional linear motion model.With the original motion path of obtained two-dimensional linear motion model estimation camera.It determines the smooth constraint condition of objective function and limits the constraint condition of original camera motion path transformation.It is directed to the solution of above-mentioned optimization problem simultaneously, obtains cutting transformation matrix.Original video sequence is converted with the transformation matrix of crop window again to obtain stable video sequence.
Description
Technical field
The present invention relates to one kind based on the estimation of camera track and the matched digital image stabilization method of characteristic block, belongs to video processing neck
Domain.
Background technique
With the development of computer and electronic communication and multimedia technology, the daily life of smart phone and computer in people
More more and more universal in work, wearable intelligent equipment has gradually become a hot spot, and the above smart machine mostly has camera function, still
During camera shooting, usually can because of the shake of photographer so that shooting video jitter so that viewer feels tired
Labor influences video effect, it is therefore desirable to carry out de-jitter to video, generate stable video.
It is most of due to the diversity of actual jitter frequency although Video Stabilization have passed through research in thirties years
Digital image stabilization method can only solve the shake of high frequency, in general smooth using motion path, but often shake the camera fortune of video
It moves data and there is the noise shaken by a small margin, such as the translation shooting or a people's progress video bat walked of handheld device
Take the photograph generated shake.
In current many Video Stabilization algorithms, Matthias Grundmann；Vivek Kwatra1 and Irfan
The Auto-Directed Video Stabilization with Robust L1 Optimal Camera that Essa is proposed
Stable optical effect after the processing video of Paths (the automatic orientation video stabilization with the powerful best video camera path L1)
Fruit is best.Objective function is constructed using L1 norm, constraint is added on camera path, stablizes video.But the algorithm is with sacrificial
The certain active movement of domestic animal, stablize after video in do not have retain script active movement.What Hui Qu, Li Song was proposed
VIDEO STABILIZATION WITH L1-L2 OPTIMIZATION (Video Stabilization of L1-L2 optimization), introduces mixing
The video stabilization algorithm of L1-L2 optimization, it is intended to eliminate unnecessary camera motion, and keep original video letter to the maximum extent
Breath.But the stability of the algorithm is not so good as the former, it is that true active movement restores and few.According to Ken-Yi Lee Yung-
The Video Stabilization using Robust that Yu Chuang Bing-Yu Chen Ming Ouhyoung is proposed
Feature Trajectories (uses the video stabilization algorithm of the Path of robustness), and the L2 norm of proposition optimizes spy
Path is levied, to retain the active movement of original photographer, so that the video after stablizing, more really.
Summary of the invention
To solve the above-mentioned problems, the present invention provides a kind of based on the estimation of camera track and the matched image stabilization of characteristic block
Method is taken the photograph then to the camera path estimated based on profession by the characteristic point camera path of consecutive frame in estimation video sequence
Camera shoots path and generates more stable video to optimize camera motion path, with further EQUILIBRIUM CALCULATION FOR PROCESS complexity and stabilization
Two importances of quality.
The present invention uses following technical scheme to solve above-mentioned technical problem:
The present invention provides a kind of based on the estimation of camera track and the matched digital image stabilization method of characteristic block, the specific steps are as follows:
Step 1, the feature that SIFT algorithm extracts each video frame in shake video sequence is converted by scale invariant feature
Point, and characteristic matching is carried out to adjacent video frames, obtain several pairs of match points；
Step 2, based on different in several pairs of match points obtained in M estimation sampling consistency MSAC algorithm removal step 1
Often point pair；
Step 3, it is fitted two-dimensional linear motion model according to the matching double points that step 2 obtains, the two dimension obtained according to fitting
Linear movement model estimates original camera path；
Step 4, the objective function of smooth paths and the constraint condition of limitation original path transformation are determined, the optimization is solved and asks
Topic, obtains the transformation matrix of crop window；
Step 5, each video frame in shake video sequence is converted based on the transformation matrix of crop window, output is steady
Fixed video sequence.
As further technical solution of the present invention, it is fitted and is shaken according to the matching double points that step 2 obtains in step 3
T frame video frame I in video sequencet-1To t-1 frame video frame ItTwo-dimensional linear motion model, further estimation obtain t
The original camera motion path O of frame video framet, Ot=F1F2…Ft。
As further technical solution of the present invention, objective function in step 4 are as follows:
Constraint condition includes:
Smoothness condition:
Intercept the window's position condition:
Wherein, α, β and γ are weight,For j-th jiao of c of crop windowjCoordinate,For 6 slack variables,
Respectively represent Ft+1Tt+1、Ft+2Tt+2、Ft+3Tt+3Middle matrix multiplication, the cutting transformation matrix T of t frame video frametParametrization to
Measure Pt=(at,bt,ct,dt,dxt,dyt)T, the cutting transformation matrix T of t+1 frame video framet+1Parametrization vector Pt+1=
(at+1,bt+1,ct+1,dt+1,dxt+1,dyt+1)T, the cutting transformation matrix T of t+2 frame video framet+2Parametrization vector Pt+2=
(at+2,bt+2,ct+2,dt+2,dxt+2,dyt+2)T, the cutting transformation matrix T of t+3 frame video framet+3Parametrization vector Pt+3=
(at+3,bt+3,ct+3,dt+3,dxt+3,dyt+3)T, at、at+1、at+2、at+3Respectively Tt、Tt+1、Tt+2、Tt+3Zooming parameter, dt、
dt+1、dt+2、dt+3Respectively Tt、Tt+1、Tt+2、Tt+3Rotation parameter, btAnd ct、bt+1And ct+1、bt+2And ct+2、bt+3And ct+3Point
It Wei not Tt、Tt+1、Tt+2、Tt+3Affine transformation parameter, dxtAnd dyt、dxt+1And dyt+1、dxt+2And dyt+2、dxt+3And dyt+3Respectively
For Tt、Tt+1、Tt+2、Tt+3Displacement parameter,W and h is respectively the width and length of video frame.
As further technical solution of the present invention, 0.9≤at,dt≤ 1.1, -0.1≤bt,ct≤0.1。
The invention adopts the above technical scheme compared with prior art, has following technical effect that
(main) purpose of the invention is to propose that one kind based on the estimation of camera track and the matched digital image stabilization method of characteristic block, is led to
Estimation camera path is crossed to generate more stable video, with two important sides of further EQUILIBRIUM CALCULATION FOR PROCESS complexity and stabilised quality
Face:
First, the match point that the present invention rejects mistake using MASC algorithm influences；
Second, the present invention is fitted the movement of camera using L2 norm, to estimate the track of camera, L2 norm can be to prevent
Only over-fitting improves the generalization ability of model.
Detailed description of the invention
Fig. 1 is flow chart of the method for the present invention.
Specific embodiment
Embodiments of the present invention are described below in detail, the example of the embodiment is shown in the accompanying drawings, wherein from beginning
Same or similar element or element with the same or similar functions are indicated to same or similar label eventually.Below by ginseng
The embodiment for examining attached drawing description is exemplary, and for explaining only the invention, and is not construed as limiting the claims.
Those skilled in the art can understand that unless otherwise defined, all terms used herein (including skill
Art term and scientific term) there is meaning identical with the general understanding of those of ordinary skill in fields of the present invention.Also
It should be understood that those terms such as defined in the general dictionary should be understood that have in the context of the prior art
The consistent meaning of meaning will not be explained in an idealized or overly formal meaning and unless defined as here.
Technical solution of the present invention is described in further detail with reference to the accompanying drawing:
A kind of method that the present invention proposes new Video Stabilization obtains shake video sequence first, video sequence is handled
At video frame one by one.SIFT (Scale-invariant feature transform) is carried out to original video sequence
Algorithm carries out characteristic point detection and matching, and the thick matching between the frame and frame of video sequence is completed according to the characteristic point detected.
By the SIFT feature of front extract and matching after, due to there is many in the SIFT feature matching estimated based on Euclidean distance
The point of erroneous matching needs to calculate using M estimation sampling consistency (M-estimator Sample and Consensus, MSAC)
Method realizes rejecting abnormalities matching double points, by obtained matching double points is fitted two-dimensional linear motion model.With what is obtained
The original motion path of two-dimensional linear motion model estimation camera.Can only smoothly high frequency be inhibited to tremble according to most of motion path
It is dynamic, low-frequency jitter cannot be handled well, and the method being fitted using motion path simulates the camera fortune that professional cameraman uses
Dynamic, camera path is by constant section, and linearity range and parabolic segment form, and by constant section, linearity range and parabolic segment form construction
Objective function, determine the smooth constraint condition of objective function and limit original camera motion path transformation constraint condition.
It is directed to the solution of above-mentioned optimization problem simultaneously, obtains the transformation matrix of crop window.The transformation matrix pair of crop window is used again
Original video sequence is converted to obtain stable video sequence.
Shake video sequence I is obtained in the present invention first, shake video sequence is processed into video frame one by one.According to
It is secondary to be denoted as I1,I2..., It(t >=1), n indicate video frame frame number.
It is carried out using scale invariant feature transfer algorithm (Scale Invariant Feature Transform, SIFT)
Feature extraction and matching: the first frame I of video sequence will be shaken first1It is rolled up with the two-dimensional Gaussian function of a variable dimension
Product operation obtains scale space, searches for the image on scale space, potential scale and choosing are identified by gaussian derivative function
Select constant characteristic point.Then on the position of each candidate, Location Scale is determined by a fitting refined model, according to
Degree of stability selected characteristic point.Then based on the gradient direction of image local, it is one or more to distribute to each characteristic point position
Direction, direction, scale and the position that subsequent all operations are all based on characteristic point are converted, to provide these features
Invariance.In the neighborhood around each characteristic point, the partial gradient of image is measured on selected scale, these gradients are become
Change a kind of expression, this deformation and the light change for indicating to allow bigger local shape into.By SIFT algorithm, is obtained
The distribution of the characteristic point of one frame.Then using SIFT algorithm to the second frame I2Characteristic point detection is carried out, and by I2The feature detected
Point and I1The characteristic point detected carries out characteristic matching and (estimates to define by Euclidean distance, with key point feature vector
Euclidean distance is measured as the similarity determination of two field pictures characteristic point, and decision threshold takes 0.8 in this patent, obtains consecutive frame
Match information.Then to third frame I3Carry out SIFT algorithm carry out characteristic point detection, then with I2Carry out Feature Points Matching.With
This analogizes, until the last frame of shake video sequence, obtains the thick match information between the consecutive frame of shake video sequence.
Due to simply estimated based on Euclidean distance SIFT feature matching in there are the point of many error hidings, these error hidings
Point will seriously affect the resolving of transformation model parameter below, it is therefore desirable to be picked the point of these error hidings using the method for robustness
It removes, this patent uses M estimation sampling consistency (M-estimator Sample and Consensus, MSAC) algorithm.MSAC
The point big with estimated model bias is considered as exceptional value by algorithm, reduces the weight with the bigger point of model bias.Pass through construction
Cost function come acquire minimum value realize reject Mismatching point pair so that shake video sequence consecutive frame between matching letter
It is more accurate to cease.Frame I is obtained to after by rejecting Mismatching pointt-1To frame ItThe two-dimensional linear motion model F of fittingt, and according to
This two-dimensional linear motion model FtEstimate the original motion path O of camerat, OtFor the original camera motion path of t frame.
The method being fitted with motion path, simulates the shooting path of professional cameraman, camera path be by constant section, it is linear
Section and parabolic segment composition, by optimization objective function, achieve the effect that smooth camera path locus.By respectively to constant
Section, linearity range and the fitting in parabolic segment path construct objective function, by minimizing objective function, reach the effect of motion compensation
Fruit.Formula is as follows:
Φ=α | | f (x) | |2+β||f2(x)||2+γ||f3(x)||2 (1)
Optimization object function is gone using L2 norm, so that Φ is minimum.f(x),f2(x) and f3(x) path fitting is respectively represented
In single order, second order and three order derivatives；α, β and γ are the weight in objective function, by the adjusting of three, reach path most
The target of optimization.
According to the camera motion that professional cameraman uses, smooth camera path is decomposed into three parts, it is respectively constant
Path, the path in constant speed path and constant acceleration.Constant path represents static camera, i.e. f (x)=0, is equivalent to phase
Machine is fixed on tripod；Constant speed path represents camera with constant speed movement, i.e. f2(x)=0, it is equivalent to camera quilt
It is placed on photographic car at the uniform velocity；Constant acceleration path is represented camera and is moved with constant acceleration, i.e. f3(x)=0 it, is equivalent to
Camera conversion process between stationary state and constant speed state.Our target is the camera after motion path fitting
Path is only made of constant path, constant speed path and constant acceleration path, has just achieved the purpose that motion compensation.
By the original motion path O of cameratWith the path X for the optimization wantedtContextual definition between the two are as follows:
Xt=OtTt (2)
Wherein, TtFor the original camera motion path O of t frametIt is transformed into the optimization camera path X of t frametTransformation square
Battle array, that is, indicate the cutting transformation matrix of t frame.
To make Φ minimum by constructing objective function, i.e., minimize f (x), f respectively2(x) and f3(x), while according to Ot
And XtRelationship and OtWith FtRelationshipIt obtains
With OtIt is known that as long as we minimize Rt=Ft+1Tt+1-Tt。
For | | f2(x)||2, available
Similarly, for | | f3(x)||2, available
TtIt can be indicated by linear movement model, using the affine matrix of 6 parameters in this patent:
Wherein, atTransformation matrix T is cut for t frametZooming parameter, dtTransformation matrix T is cut for t frametRotation ginseng
Number, btIt is that t frame cuts transformation matrix T with cttAffine transformation parameter, dxtAnd dytTransformation matrix T is cut for t frametPosition
Shifting parameter.
Enable PtTransformation matrix T is cut for t frametParametrization vector, Pt=(at,bt,ct,dt,dxt,dyt)T, it is simultaneously
The intention for guaranteeing original dither video sequence, needs to limit transformation matrix Tt, do not allow it to deviate original path, we are to parameter
Stringent boundary is arranged in the affine part changed:
0.9≤at,dt≤ 1.1, -0.1≤bt,ct≤0.1 (8)
atAnd dtLimit the variation range of scaling and rotation, btAnd ctMade by limitation tilt quantity and uneven ratio
Affine transformation has bigger rigidity.
According to the size after crop window no more than original video sequence, i.e. the 4 of crop window angle is all located at shake
In the size of video sequence,Wherein c1~c4For four angles of crop window,For cj's
Coordinate, thus the constraint of crop window are as follows:
Wherein,For the lower bound of representation parameter,The upper bound of representation parameter, at the same w and h represent video frame width and
It is long.
FtAnd TtIt can be by PtVector indicates, as a result, according to Rt=Ft+1Tt+1-Tt, can be by RtIt can be by parametrization vector PtTable
Show:
Wherein,Represent Ft+1Tt+1Middle matrix multiplication.
That is:
So minimize f (x), f2 (x) and f3It (x) is substantially exactly to minimize WithAccording to Grundmann
The thought of the linear programming of proposition is minimized by introducing N number of slack variable Wherein N is parametrization
Vector PtDimension because this patent is using affine matrix, N=6, if e be 6 slack variables parametrization to
Amount, i.e.,
Wherein,For the vector form of 6 slack variables,For 6 slack variables
For | | f (x) | |2Then have
Similarly for | | f2(x)||2, make to have if up conversion
For | | f3(x)||2, make to have if up conversion
Thus our objective function becomes:
Formula (16)~(18) become the constraint condition that optimization problem guarantees path slickness, as shown in table 1.Pass through tune
Whole α, β and γ enable rotation and translation part to obtain good transition, are unlikely to for the affine matrix part of use
There is unexpected variation.Use the ratio between value of weight parameter for 100:1 generally for affine part and translating sections.By asking
The linear programming problem of inducing diaphoresis 1 obtains the problem of optimal camera path optimal solution, obtains cutting transformation matrix Tt.Further according to obtaining
Cutting transformation matrix TtOriginal video sequence is converted to obtain stable video sequence.
Table 1
In the present embodiment, new video stabilizing method proposed by the present invention extracts characteristic point simultaneously using SIFT algorithm
It is slightly matched using characteristic point, using using M estimation sampling consistency (M-estimator Sample and
Consensus, MSAC) algorithm removal Exceptional point.By obtained matching to being fitted two-dimensional linear motion model.With
The original motion path of obtained two-dimensional linear motion model estimation camera.Determine objective function smooth constraint condition and limit
The constraint condition of original camera motion path transformation processed.It is directed to the solution of above-mentioned optimization problem simultaneously, obtains cutting transformation square
Battle array.Again original video sequence is converted to obtain stable video sequence with cutting transformation matrix.
The above, the only specific embodiment in the present invention, but scope of protection of the present invention is not limited thereto, appoints
What is familiar with the people of the technology within the technical scope disclosed by the invention, it will be appreciated that expects transforms or replaces, and should all cover
Within scope of the invention, therefore, the scope of protection of the invention shall be subject to the scope of protection specified in the patent claim.
Claims (4)
1. one kind is based on the estimation of camera track and the matched digital image stabilization method of characteristic block, which is characterized in that specific step is as follows:
Step 1, the characteristic point that SIFT algorithm extracts each video frame in shake video sequence is converted by scale invariant feature, and
Characteristic matching is carried out to adjacent video frames, obtains several pairs of match points；
Step 2, based on the abnormal point in several pairs of match points obtained in M estimation sampling consistency MSAC algorithm removal step 1
It is right；
Step 3, it is fitted two-dimensional linear motion model according to the matching double points that step 2 obtains, the two-dimensional linear obtained according to fitting
Moving description original camera path；
Step 4, the objective function of smooth paths and the constraint condition of limitation original path transformation are determined, the optimization problem is solved,
Obtain the transformation matrix of crop window；
Step 5, each video frame in shake video sequence is converted based on the transformation matrix of crop window, is exported stable
Video sequence.
2. according to claim 1 a kind of based on the estimation of camera track and the matched digital image stabilization method of characteristic block, feature exists
In being fitted to obtain t frame video frame I in shake video sequence according to the matching double points that step 2 obtains in step 3t-1To t-1
Frame video frame ItTwo-dimensional linear motion model, further estimation obtain the original camera motion path O of t frame video framet, Ot
=F1F2…Ft。
3. according to claim 1 a kind of based on the estimation of camera track and the matched digital image stabilization method of characteristic block, feature exists
In objective function in step 4 are as follows:
Constraint condition includes:
Smoothness condition:
Intercept the window's position condition:
Wherein, α, β and γ are weight,For j-th jiao of c of crop windowjCoordinate, For 6 slack variables,
Respectively represent Ft+1Tt+1、Ft+2Tt+2、Ft+3Tt+3Middle matrix multiplication, the cutting transformation matrix T of t frame video frametParametrization to
Measure Pt=(at,bt,ct,dt,dxt,dyt)T, the cutting transformation matrix T of t+1 frame video framet+1Parametrization vector Pt+1=
(at+1,bt+1,ct+1,dt+1,dxt+1,dyt+1)T, the cutting transformation matrix T of t+2 frame video framet+2Parametrization vector Pt+2=
(at+2,bt+2,ct+2,dt+2,dxt+2,dyt+2)T, the cutting transformation matrix T of t+3 frame video framet+3Parametrization vector Pt+3=
(at+3,bt+3,ct+3,dt+3,dxt+3,dyt+3)T, at、at+1、at+2、at+3Respectively Tt、Tt+1、Tt+2、Tt+3Zooming parameter, dt、
dt+1、dt+2、dt+3Respectively Tt、Tt+1、Tt+2、Tt+3Rotation parameter, btAnd ct、bt+1And ct+1、bt+2And ct+2、bt+3And ct+3Point
It Wei not Tt、Tt+1、Tt+2、Tt+3Affine transformation parameter, dxtAnd dyt、dxt+1And dyt+1、dxt+2And dyt+2、dxt+3And dyt+3Respectively
For Tt、Tt+1、Tt+2、Tt+3Displacement parameter,W and h is respectively the width and length of video frame.
4. according to claim 3 a kind of based on the estimation of camera track and the matched digital image stabilization method of characteristic block, feature exists
In 0.9≤at,dt≤ 1.1, -0.1≤bt,ct≤0.1。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910192526.7A CN110047091B (en) | 2019-03-14 | 2019-03-14 | Image stabilization method based on camera track estimation and feature block matching |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910192526.7A CN110047091B (en) | 2019-03-14 | 2019-03-14 | Image stabilization method based on camera track estimation and feature block matching |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110047091A true CN110047091A (en) | 2019-07-23 |
| CN110047091B CN110047091B (en) | 2022-09-06 |
Family
ID=67274718
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910192526.7A Active CN110047091B (en) | 2019-03-14 | 2019-03-14 | Image stabilization method based on camera track estimation and feature block matching |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110047091B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112750088A (en) * | 2020-12-16 | 2021-05-04 | 北京大学 | Method for automatically correcting and stabilizing video image based on linear programming |
| CN113050643A (en) * | 2021-03-19 | 2021-06-29 | 京东鲲鹏(江苏)科技有限公司 | Unmanned vehicle path planning method and device, electronic equipment and computer readable medium |
| CN113766132A (en) * | 2021-09-16 | 2021-12-07 | 武汉虎客影像文化传播有限公司 | Video shooting method and device |
| CN114095659A (en) * | 2021-11-29 | 2022-02-25 | 厦门美图之家科技有限公司 | Video anti-shake method, device, equipment and storage medium |
| WO2022214001A1 (en) * | 2021-04-08 | 2022-10-13 | 北京字跳网络技术有限公司 | Video image stabilization method and apparatus, electronic device, and storage medium |
| WO2022214037A1 (en) * | 2021-04-08 | 2022-10-13 | 北京字跳网络技术有限公司 | Video anti-shake processing method and apparatus, electronic device, and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150022677A1 (en) * | 2013-07-16 | 2015-01-22 | Qualcomm Incorporated | System and method for efficient post-processing video stabilization with camera path linearization |
| CN106210447A (en) * | 2016-09-09 | 2016-12-07 | 长春大学 | Video image stabilization method based on background characteristics Point matching |
| CN108564554A (en) * | 2018-05-09 | 2018-09-21 | 上海大学 | A kind of video stabilizing method based on movement locus optimization |
| CN109819158A (en) * | 2018-12-20 | 2019-05-28 | 西北工业大学 | Video image stabilization method based on optical field imaging |
-
2019
- 2019-03-14 CN CN201910192526.7A patent/CN110047091B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150022677A1 (en) * | 2013-07-16 | 2015-01-22 | Qualcomm Incorporated | System and method for efficient post-processing video stabilization with camera path linearization |
| CN106210447A (en) * | 2016-09-09 | 2016-12-07 | 长春大学 | Video image stabilization method based on background characteristics Point matching |
| CN108564554A (en) * | 2018-05-09 | 2018-09-21 | 上海大学 | A kind of video stabilizing method based on movement locus optimization |
| CN109819158A (en) * | 2018-12-20 | 2019-05-28 | 西北工业大学 | Video image stabilization method based on optical field imaging |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112750088A (en) * | 2020-12-16 | 2021-05-04 | 北京大学 | Method for automatically correcting and stabilizing video image based on linear programming |
| CN112750088B (en) * | 2020-12-16 | 2022-07-26 | 北京大学 | Method for automatically correcting and stabilizing video image based on linear programming |
| CN113050643A (en) * | 2021-03-19 | 2021-06-29 | 京东鲲鹏(江苏)科技有限公司 | Unmanned vehicle path planning method and device, electronic equipment and computer readable medium |
| WO2022214001A1 (en) * | 2021-04-08 | 2022-10-13 | 北京字跳网络技术有限公司 | Video image stabilization method and apparatus, electronic device, and storage medium |
| WO2022214037A1 (en) * | 2021-04-08 | 2022-10-13 | 北京字跳网络技术有限公司 | Video anti-shake processing method and apparatus, electronic device, and storage medium |
| CN113766132A (en) * | 2021-09-16 | 2021-12-07 | 武汉虎客影像文化传播有限公司 | Video shooting method and device |
| CN114095659A (en) * | 2021-11-29 | 2022-02-25 | 厦门美图之家科技有限公司 | Video anti-shake method, device, equipment and storage medium |
| CN114095659B (en) * | 2021-11-29 | 2024-01-23 | 厦门美图之家科技有限公司 | Video anti-shake method, device, equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110047091B (en) | 2022-09-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110047091A (en) | One kind is based on the estimation of camera track and the matched digital image stabilization method of characteristic block | |
| Shen et al. | Visual object tracking by hierarchical attention siamese network | |
| Shu et al. | Feature-metric loss for self-supervised learning of depth and egomotion | |
| US8126206B2 (en) | Image processing apparatus, image processing method, and program | |
| Xia et al. | Joint bilateral learning for real-time universal photorealistic style transfer | |
| CN110827200A (en) | Image super-resolution reconstruction method, image super-resolution reconstruction device and mobile terminal | |
| CN106210449B (en) | Multi-information fusion frame rate up-conversion motion estimation method and system | |
| CN104641398B (en) | Subject tracking device and camera | |
| US20230237771A1 (en) | Self-supervised learning method and apparatus for image features, device, and storage medium | |
| JP7412545B2 (en) | Image processing method, image processing device, and electronic equipment applying the same | |
| KR20010041862A (en) | Camera motion parameters estimation method | |
| EP3093822A1 (en) | Displaying a target object imaged in a moving picture | |
| WO2023221790A1 (en) | Image encoder training method and apparatus, device, and medium | |
| Zhang et al. | Self-supervised learning for real-world super-resolution from dual zoomed observations | |
| CN114390201A (en) | Focusing method and device thereof | |
| Buchanan et al. | Combining local and global motion models for feature point tracking | |
| Chen et al. | Generalized content-preserving warp: Direct photometric alignment beyond color consistency | |
| JP2021082118A (en) | Learning method, program, learning device, and method for manufacturing learned weight | |
| Wang et al. | Pixel-wise video stabilization | |
| Li et al. | Multi-object dynamic memory network for cell tracking in time-lapse microscopy images | |
| CN113315914B (en) | Panoramic video data processing method and device | |
| Wu et al. | Locally low-rank regularized video stabilization with motion diversity constraints | |
| Wang et al. | Adaptive video stabilization based on feature point detection and full-reference stability assessment | |
| An et al. | Rotateview: A video composition system for interactive product display | |
| Hu et al. | Digital video stabilization based on multilayer gray projection |
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 |