CN106550174A - A kind of real time video image stabilization based on homography matrix - Google Patents
A kind of real time video image stabilization based on homography matrix Download PDFInfo
- Publication number
- CN106550174A CN106550174A CN201610956105.3A CN201610956105A CN106550174A CN 106550174 A CN106550174 A CN 106550174A CN 201610956105 A CN201610956105 A CN 201610956105A CN 106550174 A CN106550174 A CN 106550174A
- Authority
- CN
- China
- Prior art keywords
- frame
- image
- layers
- video
- homography
- 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
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N5/00—Details of television systems
- H04N5/14—Picture signal circuitry for video frequency region
- H04N5/21—Circuitry for suppressing or minimising disturbance, e.g. moiré or halo
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N5/00—Details of television systems
- H04N5/14—Picture signal circuitry for video frequency region
- H04N5/144—Movement detection
- H04N5/145—Movement estimation
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Studio Devices (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of real time video image stabilization based on homography matrix, is carried out to the video sequence that the equipment such as hand-held DV, unmanned plane are obtained steady as processing.The method of the present invention comprises the steps:The angle point being evenly distributed in A, extraction video image;B, the light stream vector for calculating interframe, track the angle point of interframe movement using light stream vector;C, the angle point tracked using the algorithm correction of layering motion correction;D, the homography matrix that interframe is asked for using random sampling unification algorism;E, separates active motion compensation amount, the distortion correction amount of interframe using Kalman filter;F, homography matrix is used, motion compensation quantity and distortion correction amount carry out steady stable present frame output being obtained as conversion to video image.The present invention can effectively remove the shake of video sequence presence, and algorithm complex is relatively low, and the speed of service is fast, provides good using value for real-time processing system for video.
Description
Technical field
The invention belongs to the technical field of video information process, the present invention is a kind of video sequence based on homography matrix
The fields such as digital image stabilization method, it is adaptable to handheld device, unmanned plane.
Background technology
UAV (Unmanned Aerial Vehicle) plays more and more important effect in the field of taking photo by plane at present.Due to
Oneself motor vibration and the jiggly impact of flight path are received, the video that unmanned plane shoots inevitably has shake
Problem.These shakes have not only had a strong impact on viewing effect, and can be to the target detection of video sequence, target following and mesh
Mark not Deng post processing bring larger error.Therefore, the high dither tool in video is removed using fast and effectively Video stabilization
There is important Practical significance.
Existing Video Stabilization algorithm is broadly divided into two big class:One is the Video stabilization converted based on 2D images, and two is base
In the Video stabilization of 3D scene rebuildings.
The motion of the actual 3D of camera is reduced at the plane motion of 2D based on the Video stabilization that 2D images are converted
Reason.This kind of Video stabilization is mainly using the block feature and point feature of image carrying out the estimation of inter-frame information.It is 30FPS in frame per second
Video flowing in, the interframe time differenceRelative motion is less, using four parameters 2D similarity transformation models or six parameters it is imitative
Penetrate motion of the transformation model to interframe to be modeled, corresponding active movement amount pair is separated using the low pass filter of suitable parameter
Registering image carries out motion compensation, has certain removal effect to video jitter.But, affine mould used in 2D Video stabilizations
Type, the camera path of the 2D that scale model is estimated and the 3D camera paths non-equivalence of reality, this causes such Video stabilization pair
Obvious Video Stabilization effect is converted in visual angle and the depth of field unsatisfactory.
SFM (Struct From Motion) is used based on the Video Stabilization algorithm of 3D scene rebuildings, using 2D images
Sequence estimation goes out the 3D structures and the movable information of 3D of scene.Then the camera motion path of 3D is filtered using these information
Popin is slided, and using final smooth rear parameter compensation shake frame.This Video stabilization based on scene rebuilding is regarded for some
Angle and depth of field change significantly shake video has steady as effect well.But the algorithm is computationally intensive, parameter was calculated
Degree relies on the tracking accuracy with interframe feature, thus robustness is general.
Paper name:Content preserving warps for 3D video stabilization, periodical:ACM
Transactions on Graphics, time:2009.Liu et al. proposes a kind of based on the steady as calculating of 3D scene rebuildings
Method, and depth camera is introduced into during steady picture is processed to improve the robustness of image stabilization system.But the algorithm needs to carry out 3D to scene
Rebuild, amount of calculation is larger, it is impossible to realizes real-time processing, is difficult to apply in real-time scene.
Paper name:Calibration-free rolling shutter removal, meeting:IEEE
International Conference on Computational Photography (ICCP), time:2012.
Angle point in Grundmann et al. extraction images is used as feature, and tracks interframe angle point using optical flow method.But track algorithm pair
There is big motion in interframe, frame ambiguity effect is degenerated obvious.
For above-mentioned background content, a kind of Fast Digital Image Stabilization algorithm of effective process 3D camera path is studied, to real-time system
System is using significant.
The content of the invention
The invention aims to overcome the shortcomings of existing Video stabilization, there is provided a kind of based on the real-time of homography matrix
Video Stabilization algorithm, can effectively remove, recover the video of stable and continuous, together
When accomplish relatively low complexity and the speed of service faster, provide good using value for real-time application system.
Technical scheme:
A kind of real time video image stabilization based on homography matrix, step are as follows:
(1) obtain the frame image features point set calculating being evenly distributed to tremble using the Harris Corner Detection Algorithms of distance restraint
Current color image is first transformed into gray level image by the Harris characteristic points of dynamic video sequence image, is examined using Harris angle points
Method of determining and calculating obtains angle point, and enters row constraint with apart from r, obtains equally distributed frame image features point set;
(2) using the coordinate of the optical flow tracking algorithm keeps track next frame angle point for being layered affine constraint
1) length and width are respectively with w, the image of h sets up 5 layers of yardstick pyramid, and the corresponding length of L tomographic images is a width of
0th layer is original image yardstick, meanwhile, the corresponding angular coordinate of L layers is:
2) the interframe angle point under same yardstick is tracked in the bottom, tracking problem is converted into into inter frame image window SSD letters
Number, minimizes:
Wherein, I (x, y), J (x+ υx,y+υy) t is represented respectively1,The image at moment;(2·ωx+1)×(2·ωy+
1) represent with (px,py) centered on neighborhood window size (see Fig. 6);Pixel (px,py) elapsed timeMotion vector
(υx,υy);
Tracking problem is further converted toFinal solution obtains the corresponding angular coordinate (x of L layersL,yL);
3) affine constraint is carried out to the angle point set that the tracking of L layers is obtained, using error formula:
Wherein, FLIt is next frame L layer global homography models, its form (5);It is FL
CorrespondenceDimensional variation coefficient;Represent present frame L layer angular coordinates;Represent by with
The corresponding coordinate of next frame L layers that track is obtained;
The corresponding error threshold of L layers:
Work as εLMore than correspondence threshold value thL, using updating formula:
Initial error th is set0=3, L layers angular coordinate after correction
4) search starting point of L-1 layers isThe angle point of L-1 layers is repeated it is above-mentioned 2)-
3) tracking process, obtains the 0th layer i.e. corresponding angular coordinate of original image yardstick next frame by iterating;
(3) frame global homography matrix H before and after being calculated under original scale using RANSAC iterative algorithms0, its form (5):
(4) each amount of exercise before and after separating in the global homography matrix of frame, obtains corresponding using Kalman filter
Smooth active movement amount
1) make H '=(H0)-1, to H ' disengaging movement amounts;The translation T of amount of exercise to be separated comprising relative former frame (x,
Y), the distortion D of 2D anglecs of rotation θ, 2D scaling Z and homography matrix;
2) separation of actual motion amount is carried out using the method for four point transformation, define the length and width correspondence of original rectangle ABCD
Be frame of video length and width, area is S, obtains A ' B ' C ' D ' to rectangular transform using H ', and area is S ', now amount of zoom:
Calculate A ' B ' and C ' D ' angle thetas in the horizontal direction1And θ3, angle thetas of the A ' D ' and B ' C ' in vertical direction4And θ2,
Reduce the 2D angular errors brought by distortion with averaging method:
Because scaling and rotation can affect the calculating of translational movement, this Video stabilization is using the Z and θ for having obtained to quadrangle
A ' B ' C ' D ' compensate conversion, remove rotation and scale;Center before and after then calculating is converted respectively C (x, y), C ' (x,
Y), obtaining translational movement (see Fig. 5) is:
T(tx,ty)=C (x, y)-C ' (x, y) (10)
3) translation T, the rotation θ and scaling Z to having obtained is filtered using Kalman filter and can obtain correspondence
Active movement amount IT、IθAnd IZ, wherein for the state equation of two-dimensional vector T correspondence Kalman filter is:
Corresponding observing matrix is:
The state equation of the corresponding Kalman filter of scalar θ, Z:
Observing matrix:
(1 0 0) (14)
Wherein dt is the sampling interval;
4) above-mentioned amount of exercise is 2D, and homography conversion can bring the pattern distortion of real visual angle change and generation, it is necessary to
This distortion is corrected.Using the translation T of existing 2D, θ is rotated, scaling Z enters line translation and remove 2D to put down to A ' B ' C ' D '
Face is moved, and obtains A " B " C " D ", and A " B " C " D " only exists the distortion brought by 3D visual angle changes.The karr set up using translation T
Graceful wave filter to conversion after four apex coordinate A " B " C " D " be filtered respectively and obtain AiBiCiDi。AiBiCiDiTo ABCD tables
Show the corresponding active variable quantity of camera 3D visual angle changes, i.e., required distortion correction amount ID。
(5) the steady as conversion of nth frame is calculated using the global homography for having obtained and each smooth active movement amount
MODEL CN, image is carried out steady as conversion.
1) nth frame does homography conversion H of benchmark with N-1 framesN, 2D plane motion compensation matrix motionHN, 3D visual angles
Change corresponding distortion compensation conversion distortHN, Compensation Transformation of the nth frame with reference to N-1 frames is obtained using above-mentioned conversion:
TN=motionHN·distortHN·HN (15)
In order to the frame of video for obtaining continuous-stable is exported, using the first frame as global reference frame, therefore final nth frame is steady
As transformation model is:
CN=T1·...·TN-1·TN (16)
Using CNEnter line translation to present frame and obtain stable two field picture.
The beneficial outcomes of the present invention:
(1) interframe corner location is tracked using the optical flow tracking method for being layered affine constraint, takes full advantage of and regard
The information of frequency interframe.Meanwhile, by Layer constraint reduction transmission error between layers.With traditional unconfined light of layering
Stream tracking contrast, the optical flow tracking of the affine constraint of the multilayer in this algorithm can significantly improve interior ratio after tracking.
(2) present invention uses the homography model of 8 frees degree is modeled calculating to Inter-frame Transformation, change is not only wrapped
The translation T rotation θ scaling Z of 2D planes are contained, but also have included the image change brought by camera perspective change so that interframe
Image can be more accurately registering.
(3) present invention carries out smothing filtering to each component motion using Kalman filter, while becoming for homography
Changing the distortion produced in successive frame carries out the separation of amount of distortion, and using the active movement amount of Kalman filter point distortion,
Steady in topography is obtained better than tradition as in based on similar matrix, the Video stabilization of affine matrix.
(4) inventive algorithm complexity is only the linear function of image size, for the Video processing of 720 × 512 sizes
Speed reaches 30FPS, realizes real-time processing.
Description of the drawings
Fig. 1 is a kind of real time video image stabilization schematic flow sheet based on homography matrix.
Fig. 2 is the optical flow tracking schematic diagram for being layered affine constraint.
Fig. 3 is to whether there is distortion correction Comparative result, wherein, (a) pending frame;B () is undistorted timing result;
(c) this method distortion correction result.
Fig. 4 is the result of three kinds of digital image stabilization methods of video of taking photo by plane, and under reference frame, 21 frames and 381 frames, (a) is original respectively
Two field picture;B () is steady as result based on similarity transformation;The steady picture result of (c) L1 norm optimizations;D () is steady as result herein.
Fig. 5 is four point transformation disengaging movement amount schematic diagram.
Fig. 6 seeks poor window schematic diagram for SSD neighborhoods.
Specific embodiment
Below in conjunction with accompanying drawing and technical scheme, the specific embodiment of the present invention is further illustrated.
Embodiment
A, nth frame image is obtained from video, original angle point set is obtained using Harris Corner Detection Algorithms, for
The image of 720 × 512 yardsticks, selects constraint distance to be r=5pixel, in order to take into account the demand of real-time processing, arranges maximum angular
Points M is set to 200.
B, reading N+1 two field pictures, using the optical flow tracking algorithm keeps track nth frame angle point of Layer constraint in N+1 frames
Position, specific flow process is as shown in Figure 2.
B1, the gray level image to original N+1 two field pictures set up image pyramid, and in actual algorithm, the pyramid number of plies sets
It is set to the error threshold th of the 5, the 0th layer of affine constraint0=3pixel.Threshold value th of L layers is calculated by formula (5)L(L >
0), and using formula (2) obtain the corresponding tracking initial point of L layers.
B2, using formula (3) utilizeObtain light stream vector υopt, nth frame characteristic point is then obtained in N+1 frames
The corresponding coordinate of L layers.
B3, the matching point set using two frames in front and back in L layers calculate the global affine constraint matrix F of L layersL, utilize
Formula (4) (6) is corrected to the point set of N+1 frame L layers.
C, using the corresponding set of characteristic points of frame in front and back, the overall situation iterated to calculate by RANSAC under original image yardstick is single
Answering property matrix H0, the estimate of the exterior point ratio of wherein RANSAC is set to 0.1, and the pixel error upper limit is 3pixel, interior ratio
Lower limit be 0.25.
D, separate each component motion from global homography inverse of a matrix matrix, obtain smooth by Kalman filtering
Active movement component.
D1, calculating homography matrix H0Inverse matrix H ', H ' represent nth frame to N+1 frames kinematic matrix.
D2, one rectangle ABCD of initialization, its centre coordinate C (x, y), long width values are respectively the length and width of image.Use
H ' enters line translation to rectangle ABCD, the quadrangle A ' B ' C ' D ' after being converted, and then calculates its centre coordinate C ' (x, y).
Translation vector T (x, y) is calculated using formula (7) (8) (9), anglec of rotation θ scales Z.
D3, above three component motion is filtered using Kalman filter, is provided with the R of Kalman filter
=100, Q=0.01, set interval dt=1/30s.The active movement for obtaining three components after filtering afterwards is IT、Iθ、
IZ。
D4, using the translation T of existing 2D, rotate θ, scaling Z enters line translation to A ' B ' C ' D ' and removes 2D plane motions, obtains
To A " B " C " D ", operation is filtered using the Kalman filter model of translation filtering respectively to four summits of A ' B ' C ' D ', is obtained
To AiBiCiDi, calculate AiBiCiDiTo homography conversion I of ABCDD, IDIt is exactly required abnormal for successive frame homography conversion
Become correction matrix.
E, calculated using formula (14) (15) it is final steady as compensation matrix CN, and line translation is entered to present frame obtain surely
Settled prior image frame.
In F, example, the key frame for updating angle point is set to frame number for iFrame=10i's (i=1,2,3,4 ...)
Frame.
Claims (1)
1. a kind of real time video image stabilization based on homography matrix, it is characterised in that step is as follows:
(1) the frame image features point set being evenly distributed is obtained using the Harris Corner Detection Algorithms of distance restraint
The Harris characteristic points of shake video sequence image are calculated, current color image is transformed into into gray level image first, used
Harris Corner Detection Algorithms obtain angle point, and enter row constraint with apart from r, obtain equally distributed frame image features point set;
(2) using the coordinate of the optical flow tracking algorithm keeps track next frame angle point for being layered affine constraint
1) length and width are respectively with w, the image of h sets up 5 layers of yardstick pyramid, and the corresponding length of L tomographic images is a width of
0th layer is original image yardstick, meanwhile, the corresponding angular coordinate of L layers is:
2) the interframe angle point under same yardstick is tracked in the bottom, tracking problem is converted into into inter frame image window SSD functions, is asked
Minimum of a value:
Wherein, I (x, y), J (x+ υx,y+υy) t is represented respectively1, t1The image of+Δ t;(2·ωx+1)×(2·ωy+1)
Represent with (px,py) centered on neighborhood window size;Pixel (px,py) elapsed time Δ t motion vector (υx,υy);
Tracking problem is further converted toFinal solution obtains the corresponding angular coordinate (x of L layersL,yL);
3) affine constraint is carried out to the angle point set that the tracking of L layers is obtained, using error formula:
Wherein, FLIt is next frame L layer global homography models, its form (5);It is FLCorrespondenceDimensional variation coefficient;Represent present frame L layer angular coordinates;Represent by tracking
The corresponding coordinate of next frame L layers for arriving;
The corresponding error threshold of L layers:
Work as εLMore than correspondence threshold value thL, using updating formula:
Initial error th is set0=3, L layers angular coordinate after correction
4) search starting point of L-1 layers isRepeat above-mentioned 2) -3 to the angle point of L-1 layers)
Tracking process, obtains the 0th layer i.e. corresponding angular coordinate of original image yardstick next frame by iterating;
(3) frame global homography matrix H before and after being calculated under original scale using RANSAC iterative algorithms0, its form (5):
(4) each amount of exercise before and after separating in the global homography matrix of frame, obtains respective smoothed using Kalman filter
Active movement amount
1) make H '=(H0)-1, to H ' disengaging movement amounts;Translation T (x, y) of amount of exercise to be separated comprising relative former frame, 2D
Anglec of rotation θ, 2D scales the distortion D of Z and homography matrix;
2) separation of actual motion amount is carried out using the method for four point transformation, the length and width of the original rectangle ABCD of definition are corresponding to be
The length and width of frame of video, area are S, and A ' B ' C ' D ' are obtained to rectangular transform using H ', and area is S ', now amount of zoom:
Calculate A ' B ' and C ' D ' angle thetas in the horizontal direction1And θ3, angle thetas of the A ' D ' and B ' C ' in vertical direction4And θ2, with
Value method reduces the 2D angular errors brought by distortion:
Conversion is compensated to quadrangle A ' B ' C ' D ' using the Z and θ for having obtained, is removed rotation and is scaled;Then calculate and become
Center before and after changing is respectively C (x, y), C ' (x, y), and obtaining translational movement is:
T(tx,ty)=C (x, y)-C ' (x, y) (10)
3) translation T, the rotation θ and scaling Z to having obtained is filtered using Kalman filter and obtains corresponding active fortune
Momentum IT、IθAnd IZ, wherein for the state equation of two-dimensional vector T correspondence Kalman filter is:
Corresponding observing matrix is:
The state equation of the corresponding Kalman filter of scalar θ, Z:
Observing matrix:
(1 0 0) (14)
Wherein dt is the sampling interval;
4) above-mentioned amount of exercise is 2D, the pattern distortion that homography conversion can be brought real visual angle change and produce, to distort into
Row correction;Line translation is entered to A ' B ' C ' D ' and removes 2D plane motions using the translation T of existing 2D, rotation θ and scaling Z, obtain
A " B " C " D ", A " B " C " D " only exist the distortion brought by 3D visual angle changes;Using the Kalman filter of translation T foundation to becoming
Four apex coordinate A " B " C " D " after changing are filtered respectively and obtain AiBiCiDi;AiBiCiDiCamera 3D visual angles are represented to ABCD
Change corresponding active variable quantity, i.e., required distortion correction amount ID;
(5) the steady as transformation model of nth frame is calculated using the global homography for having obtained and each smooth active movement amount
CN, image is carried out steady as conversion
1) nth frame does homography conversion H of benchmark with N-1 framesN, 2D plane motion compensation matrix motionHN, 3D visual angle changes
Corresponding distortion compensation converts distortHN, Compensation Transformation of the nth frame with reference to N-1 frames is obtained using above-mentioned conversion:
TN=motionHN·distortHN·HN (15)
In order to the frame of video for obtaining continuous-stable is exported, using the first frame as global reference frame, therefore the steady picture of final nth frame becomes
Mold changing type is:
CN=T1·...·TN-1·TN (16)
Using CNEnter line translation to present frame and obtain stable two field picture.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610956105.3A CN106550174B (en) | 2016-10-28 | 2016-10-28 | A kind of real time video image stabilization based on homography matrix |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610956105.3A CN106550174B (en) | 2016-10-28 | 2016-10-28 | A kind of real time video image stabilization based on homography matrix |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106550174A true CN106550174A (en) | 2017-03-29 |
CN106550174B CN106550174B (en) | 2019-04-09 |
Family
ID=58393139
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610956105.3A Active CN106550174B (en) | 2016-10-28 | 2016-10-28 | A kind of real time video image stabilization based on homography matrix |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106550174B (en) |
Cited By (19)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107220935A (en) * | 2017-05-25 | 2017-09-29 | 长光卫星技术有限公司 | A kind of in-orbit video image stabilization method of video satellite |
CN107230220A (en) * | 2017-05-26 | 2017-10-03 | 深圳大学 | A kind of new space-time Harris angular-point detection methods and device |
CN107729807A (en) * | 2017-09-05 | 2018-02-23 | 南京理工大学 | Integrated external force damage prevention target identification and intelligent early-warning system |
CN108805908A (en) * | 2018-06-08 | 2018-11-13 | 浙江大学 | A kind of real time video image stabilization based on the superposition of sequential grid stream |
CN109214288A (en) * | 2018-08-02 | 2019-01-15 | 广州市鑫广飞信息科技有限公司 | It is taken photo by plane the interframe scene matching method and device of video based on multi-rotor unmanned aerial vehicle |
CN109327712A (en) * | 2018-09-18 | 2019-02-12 | 中国科学院自动化研究所 | The video of fixed scene disappears fluttering method |
CN110163818A (en) * | 2019-04-28 | 2019-08-23 | 武汉理工大学 | A kind of low illumination level video image enhancement for maritime affairs unmanned plane |
CN110337668A (en) * | 2018-04-27 | 2019-10-15 | 深圳市大疆创新科技有限公司 | Image stability augmentation method and apparatus |
CN110401784A (en) * | 2018-04-24 | 2019-11-01 | 展讯通信(天津)有限公司 | Motion smoothing method, system and the video equipment of automatic adjusument filtering strength |
CN110796010A (en) * | 2019-09-29 | 2020-02-14 | 湖北工业大学 | Video image stabilization method combining optical flow method and Kalman filtering |
CN111753856A (en) * | 2020-05-26 | 2020-10-09 | 中国人民解放军96901部队21分队 | Center self-adaptive video fast splicing method |
CN112287819A (en) * | 2020-10-28 | 2021-01-29 | 武汉三力通信有限责任公司 | High-speed multi-channel real-time image stabilizing method for video recording equipment |
CN112686204A (en) * | 2021-01-12 | 2021-04-20 | 昆明理工大学 | Video flow measurement method and device based on sparse pixel point tracking |
CN112887708A (en) * | 2021-01-22 | 2021-06-01 | 北京锐马视讯科技有限公司 | Video jitter detection method and device, video jitter detection equipment and storage medium |
CN113497861A (en) * | 2020-03-19 | 2021-10-12 | 武汉Tcl集团工业研究院有限公司 | Video stabilization method and device |
CN113689467A (en) * | 2021-07-30 | 2021-11-23 | 稿定(厦门)科技有限公司 | Feature point optimization method and system suitable for plane tracking |
CN113923369A (en) * | 2021-12-13 | 2022-01-11 | 北京海兰信数据科技股份有限公司 | Video anti-shake method and system for ship |
CN117097918A (en) * | 2023-10-19 | 2023-11-21 | 奥视(天津)科技有限公司 | Live broadcast display device and control method thereof |
WO2024051591A1 (en) * | 2022-09-09 | 2024-03-14 | 北京字跳网络技术有限公司 | Method and apparatus for estimating rotation of video, and electronic device and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101009021A (en) * | 2007-01-25 | 2007-08-01 | 复旦大学 | Video stabilizing method based on matching and tracking of characteristic |
CN101316368A (en) * | 2008-07-18 | 2008-12-03 | 西安电子科技大学 | Full view stabilizing method based on global characteristic point iteration |
CN101742122A (en) * | 2009-12-21 | 2010-06-16 | 汉王科技股份有限公司 | Method and system for removing video jitter |
CN102231792A (en) * | 2011-06-29 | 2011-11-02 | 南京大学 | Electronic image stabilization method based on characteristic coupling |
CN102427505A (en) * | 2011-09-29 | 2012-04-25 | 深圳市万兴软件有限公司 | Video image stabilization method and system on the basis of Harris Corner |
TW201523516A (en) * | 2013-12-05 | 2015-06-16 | Univ Nat Kaohsiung Applied Sci | Video frame stabilization method for the moving camera |
CN105976330A (en) * | 2016-04-27 | 2016-09-28 | 大连理工大学 | Embedded foggy-weather real-time video image stabilization method |
-
2016
- 2016-10-28 CN CN201610956105.3A patent/CN106550174B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101009021A (en) * | 2007-01-25 | 2007-08-01 | 复旦大学 | Video stabilizing method based on matching and tracking of characteristic |
CN101316368A (en) * | 2008-07-18 | 2008-12-03 | 西安电子科技大学 | Full view stabilizing method based on global characteristic point iteration |
CN101742122A (en) * | 2009-12-21 | 2010-06-16 | 汉王科技股份有限公司 | Method and system for removing video jitter |
CN102231792A (en) * | 2011-06-29 | 2011-11-02 | 南京大学 | Electronic image stabilization method based on characteristic coupling |
CN102427505A (en) * | 2011-09-29 | 2012-04-25 | 深圳市万兴软件有限公司 | Video image stabilization method and system on the basis of Harris Corner |
TW201523516A (en) * | 2013-12-05 | 2015-06-16 | Univ Nat Kaohsiung Applied Sci | Video frame stabilization method for the moving camera |
CN105976330A (en) * | 2016-04-27 | 2016-09-28 | 大连理工大学 | Embedded foggy-weather real-time video image stabilization method |
Non-Patent Citations (2)
Title |
---|
LIU FENG等: "Content-preserving warps for 3D video stabilization", 《ACM TRANSACTIONS ON GRAPHICS》 * |
Y.MATSUSHITA等: "Full-frame video stabilization with motion inpainting", 《IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
Cited By (31)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107220935A (en) * | 2017-05-25 | 2017-09-29 | 长光卫星技术有限公司 | A kind of in-orbit video image stabilization method of video satellite |
CN107220935B (en) * | 2017-05-25 | 2020-07-31 | 长光卫星技术有限公司 | Video satellite on-orbit video image stabilization method |
CN107230220A (en) * | 2017-05-26 | 2017-10-03 | 深圳大学 | A kind of new space-time Harris angular-point detection methods and device |
CN107729807A (en) * | 2017-09-05 | 2018-02-23 | 南京理工大学 | Integrated external force damage prevention target identification and intelligent early-warning system |
CN110401784A (en) * | 2018-04-24 | 2019-11-01 | 展讯通信(天津)有限公司 | Motion smoothing method, system and the video equipment of automatic adjusument filtering strength |
CN110401784B (en) * | 2018-04-24 | 2021-09-21 | 展讯通信(天津)有限公司 | Motion smoothing method and system for adaptively adjusting filtering intensity and video equipment |
CN110337668B (en) * | 2018-04-27 | 2021-08-31 | 深圳市大疆创新科技有限公司 | Image stability augmentation method and device |
CN110337668A (en) * | 2018-04-27 | 2019-10-15 | 深圳市大疆创新科技有限公司 | Image stability augmentation method and apparatus |
WO2019205087A1 (en) * | 2018-04-27 | 2019-10-31 | 深圳市大疆创新科技有限公司 | Image stabilization method and device |
CN108805908A (en) * | 2018-06-08 | 2018-11-13 | 浙江大学 | A kind of real time video image stabilization based on the superposition of sequential grid stream |
CN108805908B (en) * | 2018-06-08 | 2020-11-03 | 浙江大学 | Real-time video image stabilization method based on time sequence grid stream superposition |
CN109214288A (en) * | 2018-08-02 | 2019-01-15 | 广州市鑫广飞信息科技有限公司 | It is taken photo by plane the interframe scene matching method and device of video based on multi-rotor unmanned aerial vehicle |
CN109214288B (en) * | 2018-08-02 | 2021-08-10 | 广州市鑫广飞信息科技有限公司 | Inter-frame scene matching method and device based on multi-rotor unmanned aerial vehicle aerial video |
CN109327712A (en) * | 2018-09-18 | 2019-02-12 | 中国科学院自动化研究所 | The video of fixed scene disappears fluttering method |
CN110163818A (en) * | 2019-04-28 | 2019-08-23 | 武汉理工大学 | A kind of low illumination level video image enhancement for maritime affairs unmanned plane |
CN110796010A (en) * | 2019-09-29 | 2020-02-14 | 湖北工业大学 | Video image stabilization method combining optical flow method and Kalman filtering |
CN110796010B (en) * | 2019-09-29 | 2023-06-06 | 湖北工业大学 | Video image stabilizing method combining optical flow method and Kalman filtering |
CN113497861B (en) * | 2020-03-19 | 2022-06-24 | 武汉Tcl集团工业研究院有限公司 | Video stabilization method and device |
CN113497861A (en) * | 2020-03-19 | 2021-10-12 | 武汉Tcl集团工业研究院有限公司 | Video stabilization method and device |
CN111753856A (en) * | 2020-05-26 | 2020-10-09 | 中国人民解放军96901部队21分队 | Center self-adaptive video fast splicing method |
CN112287819A (en) * | 2020-10-28 | 2021-01-29 | 武汉三力通信有限责任公司 | High-speed multi-channel real-time image stabilizing method for video recording equipment |
CN112686204B (en) * | 2021-01-12 | 2022-09-02 | 昆明理工大学 | Video flow measurement method and device based on sparse pixel point tracking |
CN112686204A (en) * | 2021-01-12 | 2021-04-20 | 昆明理工大学 | Video flow measurement method and device based on sparse pixel point tracking |
CN112887708A (en) * | 2021-01-22 | 2021-06-01 | 北京锐马视讯科技有限公司 | Video jitter detection method and device, video jitter detection equipment and storage medium |
CN113689467A (en) * | 2021-07-30 | 2021-11-23 | 稿定(厦门)科技有限公司 | Feature point optimization method and system suitable for plane tracking |
CN113689467B (en) * | 2021-07-30 | 2022-07-12 | 稿定(厦门)科技有限公司 | Feature point optimization method and system suitable for plane tracking |
CN113923369A (en) * | 2021-12-13 | 2022-01-11 | 北京海兰信数据科技股份有限公司 | Video anti-shake method and system for ship |
CN113923369B (en) * | 2021-12-13 | 2022-03-15 | 北京海兰信数据科技股份有限公司 | Video anti-shake method and system for ship |
WO2024051591A1 (en) * | 2022-09-09 | 2024-03-14 | 北京字跳网络技术有限公司 | Method and apparatus for estimating rotation of video, and electronic device and storage medium |
CN117097918A (en) * | 2023-10-19 | 2023-11-21 | 奥视(天津)科技有限公司 | Live broadcast display device and control method thereof |
CN117097918B (en) * | 2023-10-19 | 2024-01-09 | 奥视(天津)科技有限公司 | Live broadcast display device and control method thereof |
Also Published As
Publication number | Publication date |
---|---|
CN106550174B (en) | 2019-04-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106550174A (en) | A kind of real time video image stabilization based on homography matrix | |
CN106651938B (en) | A kind of depth map Enhancement Method merging high-resolution colour picture | |
CN103761737B (en) | Robot motion's method of estimation based on dense optical flow | |
CN106251305B (en) | A kind of realtime electronic image stabilizing method based on Inertial Measurement Unit IMU | |
CN103247075B (en) | Based on the indoor environment three-dimensional rebuilding method of variation mechanism | |
Saurer et al. | Rolling shutter stereo | |
CN104318561B (en) | Method for detecting vehicle motion information based on integration of binocular stereoscopic vision and optical flow | |
US8531504B2 (en) | System and method for 3D video stabilization by fusing orientation sensor readings and image alignment estimates | |
CN112304307A (en) | Positioning method and device based on multi-sensor fusion and storage medium | |
CN107959805A (en) | Light field video imaging system and method for processing video frequency based on Hybrid camera array | |
WO2021098081A1 (en) | Trajectory feature alignment-based multispectral stereo camera self-calibration algorithm | |
CN103841297B (en) | A kind of electronic image stabilization method being applicable to resultant motion shooting carrier | |
WO2024045632A1 (en) | Binocular vision and imu-based underwater scene three-dimensional reconstruction method, and device | |
CN113721260B (en) | Online combined calibration method for laser radar, binocular camera and inertial navigation | |
CN108830925B (en) | Three-dimensional digital modeling method based on spherical screen video stream | |
CN106898048B (en) | A kind of undistorted integration imaging 3 D displaying method being suitable for complex scene | |
CN105005964A (en) | Video sequence image based method for rapidly generating panorama of geographic scene | |
CN106127696A (en) | A kind of image based on BP neutral net matching sports ground removes method for reflection | |
CN104680544B (en) | Variation scene flows method of estimation based on the regularization of 3-dimensional flow field | |
CN103544681A (en) | Non-uniform motion blurred image restoration method | |
CN109819158B (en) | Video image stabilization method based on light field imaging | |
Yuan et al. | Temporal upsampling of depth maps using a hybrid camera | |
Ramirez et al. | Open challenges in deep stereo: the booster dataset | |
CN103679680A (en) | Stereo matching method and system | |
CN113034571A (en) | Object three-dimensional size measuring method based on vision-inertia |
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 |