CN106210449A - The frame rate up-conversion method for estimating of a kind of Multi-information acquisition and system - Google Patents
The frame rate up-conversion method for estimating of a kind of Multi-information acquisition and system Download PDFInfo
- Publication number
- CN106210449A CN106210449A CN201610657029.6A CN201610657029A CN106210449A CN 106210449 A CN106210449 A CN 106210449A CN 201610657029 A CN201610657029 A CN 201610657029A CN 106210449 A CN106210449 A CN 106210449A
- Authority
- CN
- China
- Prior art keywords
- motion vector
- vector
- optical flow
- matching
- motion
- 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
- H04N5/144—Movement detection
- H04N5/145—Movement estimation
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Television Systems (AREA)
- Image Analysis (AREA)
Abstract
The present invention discloses frame rate up-conversion method for estimating and the system of a kind of Multi-information acquisition, method step is: two two field picture before and after reading, respectively it is carried out down-sampled, down-sampled image is carried out estimation based on optical flow method, original image is carried out the estimation of Block-matching, the SIFT feature of two two field pictures before and after extraction, calculate characteristic vector and carry out characteristic matching, obtain the motion vector of characteristic matching, block matching motion vector, characteristic matching motion vector and optical flow method motion vector are merged, the transmission of motion vector.The present invention compare tradition motion estimation algorithm based on Block-matching, precision improvement is obvious.For general light stream algorithm for estimating, keep that there is better performance at wisp motion vector and moving boundaries vector.
Description
Technical field
The present invention relates to video frame rate upconversion field, in particular it relates to the frame rate up-conversion fortune of a kind of Multi-information acquisition
The dynamic method and system estimated.
Background technology
Video frame rate, as the key parameter of sport video, affects the smoothness for moving object in people's watching process
Sexual experience.In low frame-rate video, the pixel qualities of adjacent interframe fluctuates and is easily caused film flicker sense (flickering), and
Have in moving scene or the sports cast of the same direction of motion, be then readily observed motion jitter (motion
And animation sense jittering).High frame-rate video can be substantially improved these visual experiences, brings more preferable comprehensive visual quality
With picture fluency, become developing direction and the important need of future video application.
Frame rate up-conversion FRUC is as an important technology of Video post-processing, at Computer Vision and DTV etc.
Field existing numerous studies work and related ends.At present, main flow practicality FRUC method is mainly by video sequence consecutive frame
Between dependency, carry out interpolation along movement locus of object, be therefore also referred to as frame rate up-conversion based on motion compensation
(Motion compensated FRUC,MC-FRUC).Existing research is mainly around the estimation of frame rate up-conversion and benefit
Repaying interpolation the two key point, the accuracy just improving motion vector field and the reliability strengthening interpolation image have obtained a series of
Achievement in research.According to the producing method of motion vector, can be divided into based on Block-matching, characteristic matching, region, gradient or conversion
Territory etc. five class commonly uses method for estimating.Wherein the method for Block-matching efficiently easily realizes with hardware due to simple, in practice
To being widely applied.But the flatness of block matching method is poor, especially under dull background, the motion of Block-matching is vowed
Amount noise is the biggest.The most most block matching method is difficult to accurately estimation, the most in practice, base for big motion vector
Often easily there is mistake on motion vector boundary and dynamic background in method in Block-matching, reduces interpolation frame video
Visual quality.For Block-matching produced problem, research worker proposes a variety of method, and the motion including multiresolution is estimated
Meter, indirectly or directly applies smoothness constraint and promotes picture quality further.But nonetheless, the quality of interleave image
Still need to be improved further.
Optical flow method, the variation optical flow method being based especially on Horn-Schunck had obtained further sending out in the last few years
Exhibition." the High Accuracy Optical Flow Estimation Based on that wherein Brox et al. delivers on ECCV
A Theory for Warping " method is widely used in actual applications.This method utilizes the framework of layering, with
Vector field is smoothed by the mode of Shi Liyong global optimization.Experiment finds, the vector field that optical flow method produces is no matter in precision
Or flatness is all compared the method for Block-matching and is obviously improved, and can be applied in the estimation of frame rate up-conversion.
But also there is its weak point in traditional optical flow method.First it is that traditional optical flow method uses the framework being layered, the most permissible
The convergence of accelerated motion vector, but it is easily lost the motion vector for wisp in the superiors and describes.This situation is little
The size of object compare with its motion vector gap little when particularly evident.Additionally, due to the existence of overall situation smoothness constraint,
Optical flow method is inaccurate for the estimation of big motion vector, and easily goes out easily to occur mistake near big object of which movement vector
Assemble.
The method of feature extraction and characteristic matching is widely used in image procossing, wherein SIFT feature,
SURF feature uses more.SIFT feature has following characteristics, be first SIFT feature be the local feature of image, it is to rotation
Turn, scaling, brightness flop maintain the invariance, and visual angle change, affine transformation, noise are also kept a certain degree of stable
Property;Secondly, SIFT feature uniqueness is good, informative, it is adaptable to carry out fast and accurately in magnanimity property data base
Join;Additionally SIFT has volume, even if several objects of minority can also extract substantial amounts of SIFT feature vector.SIFT feature
Method mainly comprises the following steps, is first metric space extremum extracting, and key point positions, and direction determines, key point describes.
By the key point of two continuous frames image being calculated and mating, the motion of corresponding key point between two frames can be obtained
Vector.Due to the characteristic of SIFT feature, the motion vector obtained by characteristic matching is for compound movement and the arrow that moves greatly
Amount all has well estimates performance.But SIFT feature is the most sparse and the position of key point is unpredictable.Additionally, due to
Being Point matching, seldom or easily there is mistake in texture simple region in the motion vector that SIFT feature coupling obtains.
Through retrieval, Publication No. CN104915966A, the Chinese invention application of Application No. CN201510233587.5, it is somebody's turn to do
A kind of frame rate up-conversion method for estimating based on Kalman filtering of disclosure of the invention and system.Described method includes following step
Rapid: first parameter and the state initial value of Kalman filter model are set so that model coincide with real system;Then pass through
First carry out one-way movement estimation, after be mapped to the strategy of interpolation frame and obtain motion vector observation;When finally using a kind of gain
Measurement vector is updated by the kalman filter method become, thus obtains motion vector the most accurately.
But the method for estimating of this kind of frame rate up-conversion is estimated for wisp motion vector and big motion vector
Count not accurate enough.The most sane for compound movement vector.
Summary of the invention
For the defect of prior art, it is an object of the invention to provide estimation side in the frame rate up-conversion of a kind of video
Method and system, by merging by optical flow method motion vector, characteristic matching motion vector and block matching motion vector,
In conjunction with motion post processing means, thus provide high performance motion vector, improve the visual quality of interpolation frame
According to the first aspect of the invention, it is provided that the frame rate up-conversion method for estimating of a kind of Multi-information acquisition, described
Method comprises the steps:
Step one: read two continuous frames image from high definition video steaming;
Step 2: the view data of step one is carried out down-sampled;
Step 3: two frame data before and after in step 2 are carried out estimation based on optical flow method, obtains optical flow method motion
Vector;
Step 4: two frame image datas before and after in step one are carried out estimation based on Block-matching, obtains Block-matching
Motion vector;
Step 5: two frame image datas before and after in step one are carried out SIFT feature extraction, and calculates each feature
Characteristic vector, and carry out characteristic matching, obtain characteristic of correspondence motion vector;
Step 6: the characteristic kinematic vector step 3 that the block matching motion vector that step 4 obtained, step 5 obtain
The optical flow method motion vector obtained merges, the motion vector field after being merged;
Step 7: the motion vector field after the fusion obtaining step 6 uses the mode of reliable motion vectors transmission to carry out
Revise;
Step 8: return to step one, reads lower two two field pictures.
Preferably, in step 2, the method for image drop sampling is that level, every 4 of vertical direction take a point.
Preferably, in step 3, optical flow method uses classical variation optical flow method.
Preferably, in step 4, block matching algorithm uses 3-dimensional recursive search.
Preferably, in step 6, characteristic kinematic vector, block matching motion vector vector merge with optical flow method motion vector,
Step is:
S1: obtain the optical flow method motion vector field vo that obtained by optical flow method, characteristic kinematic vector that characteristic matching obtains
Vf and block matching motion vector field vb, create vector field vm after merging;
S2: according to from left to right, order from top to bottom scans the motion vector field vm after fusion successively;
S3: judge in current position, characteristic matching motion vector field vf whether existing characteristics motion vector, if do not deposited
, forward S4 to, otherwise forward S5 to;
S4: computing block match motion vector field vb at the margin of image element SADB of current position respective motion vectors, and
Optical flow method motion vector field vo is at the margin of image element SADO of current position respective motion vectors;
S5: select motion vector corresponding to minima as vector field vm after merging in present bit from SADB with SADO
Put the motion vector at place;Forward S8 to;
S6: calculate the optical flow method motion vector field vo margin of image element SADO in current position respective motion vectors respectively,
And characteristic kinematic vector vf exists in the margin of image element SADF of current position respective motion vectors, block matching motion vector field vb
The margin of image element SADB of current position respective motion vectors;
S7: the motion vector selecting minima corresponding from SADB, SADO with SADF exists as vector field vm after merging
The motion vector of current position;Forward S8 to;
S8: scanned the motion vector field vm after fusion, the most then exit and obtain motion vector field vm, otherwise
Return S2.
It is highly preferred that in S4 and S6, the method for the margin of image element calculating motion vector is:
Obtaining current pixel location F (x), pixel RF (x+v) of reference frame, wherein x is current coordinate position, and v is to work as
The motion vector that front position is corresponding, RF is reference frame;
SAD=| F (x) RF (x+v) |, wherein SAD is required margin of image element.
Preferably, in step 7, motion vector transmits, and step is:
Obtain the optical flow method motion vector field vo that obtained by optical flow method, characteristic kinematic vector vf that characteristic matching obtains with
And block matching motion vector field vb, if creating vector field vm after merging;
S701: according to from left to right, order from top to bottom scans vector field vm after fusion successively;
S702: for current location (px, py), wherein px is horizontal position coordinate, and py is vertical position coordinate, vm (px,
Py) it is the vector of current location, and obtains leftward position vector vm (px-1, py) successively, upper position vector vm (px, py-
1), the vector vm (px-1, py-1) of top-left position;
S703: the margin of image element of frame before and after four vector correspondences in calculating S702, selects minimum margin of image element corresponding
Vector the vector field vm vector in current location after merging is updated;
S704: scanned full frame images, if it is not, return S701, otherwise enters S705;
S705: according to from right to left, order from top to bottom scans vector field vm after fusion successively;
S706: for current location (px, py), wherein px is horizontal position coordinate, and py is vertical position coordinate, vm (px,
Py) it is the vector of current location, and obtains right positions vector vm (px+1, py) successively, lower position vector vm (px, py+
1), the vector vm (px+1, py+1) of bottom-right location;
S707: the margin of image element of frame before and after four vector correspondences in calculating S706, selects minimum margin of image element corresponding
Vector vm is updated at the vector of current location;
708: scanned full frame images, if it is not, return S705, otherwise exit.
According to the second aspect of the invention, it is provided that the frame rate up-conversion movement estimation system of a kind of Multi-information acquisition, including:
Image input module: read two continuous frames image from high definition video steaming;
Down-sampled module: the view data of image input module input is carried out down-sampled;
Optical flow method motion vector obtains module: two frame data before and after after down-sampled module are carried out fortune based on optical flow method
Dynamic estimation, obtains optical flow method motion vector;
Block matching motion vector obtains module: before and after inputting image input module, two frame image datas are carried out based on block
The estimation of coupling, obtains block matching motion vector;
Characteristic kinematic vector obtains module: before and after inputting image input module, two frame image datas carry out SIFT feature
Extract, and calculate the characteristic vector of each feature, and carry out characteristic matching, obtain characteristic of correspondence motion vector;
Motion vector field Fusion Module: block matching motion vector that above-mentioned each module is obtained, characteristic kinematic vector light
Stream method motion vector merges, the motion vector field after being merged;
Vector transfer module: the motion vector field after the fusion obtain motion vector field Fusion Module uses reliable movement
The mode of vector transmission is modified.
Compared with prior art, the present invention has a following beneficial effect:
The present invention, because have employed the fusion method of much information, is especially considering that the method for characteristic matching is for big fortune
Dynamic vector and complicated vector there is well sign, and block matching motion method has for wisp and preferably estimates
Meter, in terms of estimation, the invention performance than Publication No. CN104915966A is good.
Further, the present invention:
1. by using optical flow method to can ensure that the optical flow method motion vector field obtained is essentially smooth;
2. estimation based on Block-matching, for ensureing that the precision of the motion vector of wisp has positive effect;
3. the characteristic kinematic vector of characteristic matching, has well estimation for big motion vector and compound movement vector
Effect, it is possible to be obviously improved frame rate up-conversion performance under complex scene;
4. the advantage making full use of each method for estimating in fusion process, carries out mutual mutual, right by judging
Answer the multiple motion vector in position corresponding margin of image element, select current position optimal motion vector, promote entirety vow
The reliability of amount;
5. the transmission of motion vector is to select current location optimum vector according to surrounding vectors, on the one hand can be by correct
Vector transmits, and additionally also ensure that the flatness of overall motion vector field;
6. the transmission of motion vector uses the motion transmission of twice different directions, can sufficiently be transmitted by correct vector,
Avoid the occurrence of omission.
Accompanying drawing explanation
By the detailed description non-limiting example made with reference to the following drawings of reading, the further feature of the present invention,
Purpose and advantage will become more apparent upon:
Fig. 1 is the method flow diagram of one embodiment of the invention;
Fig. 2 is the motion vector transmission schematic diagram of one embodiment of the invention;
Fig. 3 is one embodiment of the invention system architecture diagram.
Detailed description of the invention
Below in conjunction with specific embodiment, the present invention is described in detail.Following example will assist in the technology of this area
Personnel are further appreciated by the present invention, but limit the present invention the most in any form.It should be pointed out that, the ordinary skill to this area
For personnel, without departing from the inventive concept of the premise, it is also possible to make some deformation and improvement.These broadly fall into the present invention
Protection domain.
As it is shown in figure 1, be an embodiment flow chart of the frame rate up-conversion method for estimating of Multi-information acquisition of the present invention,
Concrete, this embodiment comprises the steps:
1 reads the two continuous frames image in video sequence, and is stored as f1, f2 respectively;
2 image drop samplings
For HD video (resolution is 1920x1080), in order to reduce follow-up computation complexity, carried out level
Down-sampled with vertical direction.Every four point selection of horizontal direction wherein, the most every four point selection therein
Point as the image after down-sampled, can obtain in this manner down-sampled after image, f1_s and f2_s, its resolution is divided
Wei 480x270;For other resolution, can select in the case of in view of motion estimation performance and computing capability
Select suitable down-sampled mode.
3 optical flow method estimation obtain the light stream vector field of correspondence.
The estimation of optical flow method can use prior art to realize, and such as uses the paper that Brox delivers on ECCV
Side in " High Accuracy Optical Flow Estimation Based on a Theory for Warping "
Method;Wherein input picture be down-sampled after image f1_s and f2_s, the motion vector obtained points to before f2_s from f1_s
To motion vector vo;
4 block-based motion estimation obtain block matching motion vector field.
Block matching method uses the method for three-dimensional recursive search, specifically:
The two continuous frames image of 4.1 inputs is f1, f2;
Whole two field picture is carried out dividing according to the block of NxN size by 4.2, and the size of N should be the fall with image here
The yardstick of sampling is consistent.In the present embodiment, N=4 is taken;
4.3 according to from left to right, and scanning sequency from top to bottom travels through the block of each NxN size successively from the upper left corner;
4.4 are positioned at for position that (x, y) block at place, its candidate vector comes from the vector being computed, and is i.e. working as
Candidate vector at front position come from left position (x-1, y) the motion vector vleft at place, upper position (x, y-1) place
Motion vector vabove, and zero vector v0.Calculate the sad value of each candidate motion vector.Wherein for vector vt, level
Direction motion vector is vtx, and movement in vertical direction vector is vty.Its sad value is equal to f1 (x, y)-f2 (x+vtx, y+vty)
Absolute value.Wherein (x y) represents, with f2 (x+vtx, y+vty), the block that size is NxN to f1;Therefrom select sad value minimum corresponding
Motion vector, is defined as vc;
4.5 are respectively vcx and vcy for the motion vector vc obtained, the value of its horizontal and vertical.The motion updated is waited
Selecting vector is (vcx-1, vcy), (vcx+1, vcy), (vcx, vcy-1), (vcx, vcy+1).Side according to above-mentioned middle calculating SAD
Formula calculates the SAD that vector is corresponding respectively, and the vector that wherein sad value is minimum is as final block matching motion vector;
4.6 scan whole two field picture, obtain block matching motion vector field vb.
5 feature extractions, characteristic matching and calculating characteristic kinematic vector field.
The feature used is SIFT feature, specifically:
The two continuous frames image of 5.1 inputs is f1, f2;
5.2 extract image f1, the characteristic point on f2 respectively, are designated as k1, k2;
5.3 couples of k1, k2 mate, and obtain characteristic of correspondence match motion vector field v;
5.4 in the same size, to vector field v according to front in order to block matching motion vector field and light stream motion vector field
The down-sampled mode stated carries out down-sampled, obtains vector field vf.
6 characteristic kinematic vectors, block matching motion vector optical flow method motion vector merge.Specifically:
The 6.1 light stream vector field vo reading a size of 480x270, Block-matching vector field vb and characteristic matching vector field vf,
And set up vector field vm after the fusion of comparable size;
6.2 according to from left to right, and scanning sequency from top to bottom, from the beginning of upper left, block-by-block scans;
6.3 current scan positions are p.Because the motion vector of characteristic matching is sparse, it is therefore desirable to judge current
Position, whether characteristic matching vector field vf exists the motion vector of correspondence.If it does not, perform step 6.4, otherwise perform
Step 6.5;
6.4 computing block match motion vector vb, optical flow method motion vector vo is at absolute value corresponding to current position vector
Error.The present embodiment uses absolute error minimum and as judgment criterion herein, but those skilled in the art is easy to understand
The judgment criterion of the present embodiment can also choose other, as least mean-square error and etc.;Assuming that motion vector is v, absolute value is by mistake
Difference sad=| f (p) f (p+v) |.The vector that absolute value error is minimum, for the optimal motion vector of current location, is stored in vm
Correspondence position;
6.5 computing block match motion vector vb, optical flow method motion vector vo and characteristic matching vf are vowed in current position
The absolute value error that amount is corresponding.Absolute value error is identical with the definition in 6.4.The motion vector of corresponding least absolute value error is
Optimal motion vector, the correspondence position being stored in vm;
6.6 scan whole positions, the motion vector field vm after being merged.
7 motion vector transmission.
The purpose of motion vector transmission is to be diffused by correct motion vector, the precision of lifter motion vector field, and
And ensure the flatness of overall vector field.The transmission of motion vector can carry out the transmission of twice distinct methods.As shown in Figure 2.
7.1 according to from left to right, and order from top to bottom scans fusion vector field vm successively;
7.2 for current location p, and the coordinate of its horizontal and vertical is respectively px, py, and corresponding motion vector is v0.Its
In its left position (px-1, py) vector v1, upper position (px, py-1) motion vector v2, top-left position (px-1, py-1)
Motion vector v3.According to current scan process order, v1, v2 and v3 are the vectors that scan process is crossed;
7.3 calculate v0, v1, v2 and v3 respective pixel value difference, here can be with absolute value error or other similar mark
Accurate.Vx, vy are respectively for vector v, horizontal displacement and vertical displacement, need to calculate its position corresponding on the original image,
And also to increase robustness, use the margin of image element that before and after calculating, two frames block on movement locus is corresponding.Wherein fall is adopted
Coordinate after sample is px, py respectively.The extending space N=2 used;So horizontal coordinate model of corresponding blocks on original resolution
Enclose h and be equal to [-N+4*py, N+4* (py+1)-1] equal to [-N+4*px, N+4* (px+1)-1], the coordinate range v of vertical direction.
In the present embodiment, margin of image element use absolute value error sad as standard, sad=| f1 (h, v) f2 (h+vx, v+vy) |.
From v0, v1, v2, v3, select vector minimum for sad, as current optimal motion vector, and update fusion vector field vm;
7.4 have scanned the whole positions merging vector field vm;
7.5 according to from top to bottom, and position from right to left scans the most updated motion vector field vm successively;
7.6 for current location p, and the coordinate of its horizontal and vertical is respectively px, py, and corresponding motion vector is v0.Its
In its location right (px+1, py) vector v1, lower position (px, py+1) motion vector v2, bottom-right location (px+1, py+1)
Motion vector v3.According to current scan process order, v1, v2 and v3 are the vectors that scan process is crossed;
7.7 is identical with 7.3, calculates v0 respectively, the margin of image element that v1, v2, v3 are corresponding.Select the motion that margin of image element is minimum
Vector, as merging the vector field vm optimal motion vector in current location;
7.8 have scanned the whole positions merging vector field vm.
As it is shown on figure 3, describe based on said method, it is provided that become in the frame per second of a kind of Multi-information acquisition realizing said method
Change movement estimation system, including:
Image input module: read two continuous frames image from high definition video steaming;
Down-sampled module: the view data of image input module input is carried out down-sampled;
Optical flow method motion vector obtains module: two frame data before and after after down-sampled module are carried out fortune based on optical flow method
Dynamic estimation, obtains optical flow method motion vector;
Block matching motion vector obtains module: before and after inputting image input module, two frame image datas are carried out based on block
The estimation of coupling, obtains block matching motion vector;
Characteristic kinematic vector obtains module: before and after inputting image input module, two frame image datas carry out SIFT feature
Extract, and calculate the characteristic vector of each feature, and carry out characteristic matching, obtain characteristic of correspondence motion vector;
Motion vector field Fusion Module: block matching motion vector that above-mentioned each module is obtained, characteristic kinematic vector light
Stream method motion vector merges, the motion vector field after being merged;
Vector transfer module: the motion vector field after the fusion obtain motion vector field Fusion Module uses reliable movement
The mode of vector transmission is modified.
In the frame rate up-conversion method for estimating step of the technology that above-mentioned each module implements and Multi-information acquisition
Technology is identical, does not repeats them here.
The present invention uses the mode of Multi-information acquisition to carry out the estimation of frame rate up-conversion, utilizes much information each
Advantage, thus the accuracy of lifter motion vector field.The present invention is for ensureing the slickness of motion vector field, simultaneously for carrying
Rise, for the estimation of wisp motion vector and the estimation of complex scene, there is remarkable result.
Although present disclosure has been made to be discussed in detail by above preferred embodiment, but it should be appreciated that above-mentioned
Description is not considered as limitation of the present invention.After those skilled in the art have read foregoing, for the present invention's
Multiple amendment and replacement all will be apparent from.Therefore, protection scope of the present invention should be limited to the appended claims.
Claims (7)
1. the frame rate up-conversion method for estimating of a Multi-information acquisition, it is characterised in that comprise the steps:
Step one: read two continuous frames image from high definition video steaming;
Step 2: the view data of step one is carried out down-sampled;
Step 3: two continuous frames data in step 2 are carried out estimation based on optical flow method, obtains optical flow method motion vector;
Step 4: two continuous frames view data in step one is carried out estimation based on Block-matching, obtains block matching motion
Vector;
Step 5: two continuous frames view data in step one is carried out SIFT feature extraction, and calculates the feature of each feature
Vector, and carry out characteristic matching, obtain characteristic of correspondence motion vector;
Step 6: the characteristic kinematic vector step 3 that the block matching motion vector that step 4 obtained, step 5 obtain obtains
Optical flow method motion vector merge, the motion vector field after being merged;
Step 7: the motion vector field after the fusion obtaining step 6 uses the mode of reliable motion vectors transmission to repair
Just;
Step 8: return to step one, reads lower two two field pictures.
Method the most according to claim 1, it is characterised in that in step 2, the method for image drop sampling is level, vertical
Every 4 of direction takes a point.
Method the most according to claim 1, it is characterised in that in step 4, block matching algorithm uses 3-dimensional recurrence and searches
Rope.
Method the most according to claim 1, it is characterised in that in step 6, characteristic kinematic vector, block matching motion vector
Vector merges with optical flow method motion vector, and step is:
S1: obtain the optical flow method motion vector field vo that obtained by optical flow method, characteristic kinematic vector vf that characteristic matching obtains with
And block matching motion vector field vb, create vector field vm after merging;
S2: according to from left to right, order from top to bottom scans the motion vector field vm after fusion successively;
S3: judge in current position, characteristic matching motion vector field vf whether existing characteristics motion vector, if do not existed, turns
To S4, otherwise forward S5 to;
S4: computing block match motion vector field vb is at the margin of image element SADB of current position respective motion vectors, and light stream
Method motion vector field vo is at the margin of image element SADO of current position respective motion vectors;
S5: select motion vector corresponding to minima as vector field vm after merging in current position from SADB with SADO
Motion vector;Forward S8 to;
S6: calculate the optical flow method motion vector field vo margin of image element SADO in current position respective motion vectors respectively, and
Characteristic kinematic vector vf is at the margin of image element SADF of current position respective motion vectors, and block matching motion vector field vb is currently
The margin of image element SADB of position respective motion vectors;
S7: select motion vector corresponding to minima as vector field vm after merging currently from SADB, SADO with SADF
The motion vector of position;Forward S8 to;
S8: scanned the motion vector field vm after fusion, the most then exit and obtain motion vector field vm, otherwise returning
S2。
Method the most according to claim 4, it is characterised in that in S4 and S6, calculates the side of the margin of image element of motion vector
Method is:
Obtaining current pixel location F (x), pixel RF (x+v) of reference frame, wherein x is current coordinate position, and v is present bit
Putting the motion vector of correspondence, RF is reference frame;
SAD=| F (x) RF (x+v) |, wherein SAD is required margin of image element.
6., according to the method described in any one of claim 1-5, it is characterised in that in step 7, motion vector transmits, step
For:
Obtain the optical flow method motion vector field vo that obtained by optical flow method, characteristic kinematic vector vf that characteristic matching obtains and block
Match motion vector field vb, if creating vector field vm after merging;
S701: according to from left to right, order from top to bottom scans vector field vm after fusion successively;
S702: for current location (px, py), wherein px is horizontal position coordinate, and py is vertical position coordinate, vm (px, py)
For the vector of current location, and obtain leftward position vector vm (px-1, py) successively, upper position vector vm (px, py-1), left
The vector vm (px-1, py-1) of upper position;
S703: the margin of image element of frame before and after four vector correspondences in calculating S702, selects the arrow that minimum margin of image element is corresponding
Measure and the vector field vm vector in current location after merging is updated;
S704: scanned full frame images, if it is not, return S701, otherwise enters S705;
S705: according to from right to left, order from top to bottom scans vector field vm after fusion successively;
S706: for current location (px, py), wherein px is horizontal position coordinate, and py is vertical position coordinate, vm (px, py)
For the vector of current location, and obtain right positions vector vm (px+1, py) successively, lower position vector vm (px, py+1), right
The vector vm (px+1, py+1) of lower position;
S707: the margin of image element of frame before and after four vector correspondences in calculating S706, selects the arrow that minimum margin of image element is corresponding
Vm is updated by amount at the vector of current location;
708: scanned full frame images, if it is not, return S705, otherwise exit.
7. realizing a frame rate up-conversion movement estimation system for the Multi-information acquisition of claim 1-6 any one method, it is special
Levy and be to include:
Image input module: read two continuous frames image from high definition video steaming;
Down-sampled module: the view data of image input module input is carried out down-sampled;
Optical flow method motion vector obtains module: two frame data before and after after down-sampled module are carried out motion based on optical flow method and estimates
Meter, obtains optical flow method motion vector;
Block matching motion vector obtains module: before and after inputting image input module, two frame image datas are carried out based on Block-matching
Estimation, obtain block matching motion vector;
Characteristic kinematic vector obtains module: two frame image datas before and after image input module input are carried out SIFT feature and carries
Take, and calculate the characteristic vector of each feature, and carry out characteristic matching, obtain characteristic of correspondence motion vector;
Motion vector field Fusion Module: block matching motion vector that above-mentioned each module is obtained, characteristic kinematic vector optical flow method
Motion vector merges, the motion vector field after being merged;
Vector transfer module: the motion vector field after the fusion obtain motion vector field Fusion Module uses reliable motion vectors
The mode of transmission is modified.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610657029.6A CN106210449B (en) | 2016-08-11 | 2016-08-11 | Multi-information fusion frame rate up-conversion motion estimation method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610657029.6A CN106210449B (en) | 2016-08-11 | 2016-08-11 | Multi-information fusion frame rate up-conversion motion estimation method and system |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106210449A true CN106210449A (en) | 2016-12-07 |
CN106210449B CN106210449B (en) | 2020-01-07 |
Family
ID=57514586
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610657029.6A Active CN106210449B (en) | 2016-08-11 | 2016-08-11 | Multi-information fusion frame rate up-conversion motion estimation method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106210449B (en) |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107316275A (en) * | 2017-06-08 | 2017-11-03 | 宁波永新光学股份有限公司 | A kind of large scale Microscopic Image Mosaicing algorithm of light stream auxiliary |
CN108574844A (en) * | 2017-03-13 | 2018-09-25 | 信阳师范学院 | A kind of more tactful video frame rate method for improving of space-time remarkable perception |
CN108833920A (en) * | 2018-06-04 | 2018-11-16 | 四川大学 | A kind of DVC side information fusion method based on light stream and Block- matching |
WO2019191889A1 (en) * | 2018-04-02 | 2019-10-10 | 北京大学 | Method and device for video processing |
CN110446107A (en) * | 2019-08-15 | 2019-11-12 | 电子科技大学 | A kind of video frame rate upconversion method suitable for scaling movement and light and shade variation |
CN110555805A (en) * | 2018-05-31 | 2019-12-10 | 杭州海康威视数字技术股份有限公司 | Image processing method, device, equipment and storage medium |
CN111277863A (en) * | 2018-12-05 | 2020-06-12 | 阿里巴巴集团控股有限公司 | Optical flow frame interpolation method and device |
CN111405316A (en) * | 2020-03-12 | 2020-07-10 | 北京奇艺世纪科技有限公司 | Frame insertion method, electronic device and readable storage medium |
CN111741304A (en) * | 2019-03-25 | 2020-10-02 | 四川大学 | Method for combining frame rate up-conversion and HEVC (high efficiency video coding) based on motion vector refinement |
CN112511859A (en) * | 2020-11-12 | 2021-03-16 | Oppo广东移动通信有限公司 | Video processing method, device and storage medium |
CN112954454A (en) * | 2021-02-08 | 2021-06-11 | 北京奇艺世纪科技有限公司 | Video frame generation method and device |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103325108A (en) * | 2013-05-27 | 2013-09-25 | 浙江大学 | Method for designing monocular vision odometer with light stream method and feature point matching method integrated |
CN105023278A (en) * | 2015-07-01 | 2015-11-04 | 中国矿业大学 | Movable target tracking method and system based on optical flow approach |
CN105488812A (en) * | 2015-11-24 | 2016-04-13 | 江南大学 | Motion-feature-fused space-time significance detection method |
CN105590327A (en) * | 2014-10-24 | 2016-05-18 | 华为技术有限公司 | Motion estimation method and apparatus |
-
2016
- 2016-08-11 CN CN201610657029.6A patent/CN106210449B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103325108A (en) * | 2013-05-27 | 2013-09-25 | 浙江大学 | Method for designing monocular vision odometer with light stream method and feature point matching method integrated |
CN105590327A (en) * | 2014-10-24 | 2016-05-18 | 华为技术有限公司 | Motion estimation method and apparatus |
CN105023278A (en) * | 2015-07-01 | 2015-11-04 | 中国矿业大学 | Movable target tracking method and system based on optical flow approach |
CN105488812A (en) * | 2015-11-24 | 2016-04-13 | 江南大学 | Motion-feature-fused space-time significance detection method |
Non-Patent Citations (1)
Title |
---|
吴垠、李良福等: "基于尺度不变特征的光流法目标跟踪技术研究", 《计算机工程与应用》 * |
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108574844B (en) * | 2017-03-13 | 2021-09-28 | 信阳师范学院 | Multi-strategy video frame rate improving method for space-time significant perception |
CN108574844A (en) * | 2017-03-13 | 2018-09-25 | 信阳师范学院 | A kind of more tactful video frame rate method for improving of space-time remarkable perception |
CN107316275A (en) * | 2017-06-08 | 2017-11-03 | 宁波永新光学股份有限公司 | A kind of large scale Microscopic Image Mosaicing algorithm of light stream auxiliary |
WO2019191889A1 (en) * | 2018-04-02 | 2019-10-10 | 北京大学 | Method and device for video processing |
CN110555805A (en) * | 2018-05-31 | 2019-12-10 | 杭州海康威视数字技术股份有限公司 | Image processing method, device, equipment and storage medium |
CN110555805B (en) * | 2018-05-31 | 2022-05-31 | 杭州海康威视数字技术股份有限公司 | Image processing method, device, equipment and storage medium |
CN108833920A (en) * | 2018-06-04 | 2018-11-16 | 四川大学 | A kind of DVC side information fusion method based on light stream and Block- matching |
CN108833920B (en) * | 2018-06-04 | 2022-02-11 | 四川大学 | DVC side information fusion method based on optical flow and block matching |
CN111277863A (en) * | 2018-12-05 | 2020-06-12 | 阿里巴巴集团控股有限公司 | Optical flow frame interpolation method and device |
CN111277863B (en) * | 2018-12-05 | 2022-06-14 | 阿里巴巴集团控股有限公司 | Optical flow frame interpolation method and device |
CN111741304A (en) * | 2019-03-25 | 2020-10-02 | 四川大学 | Method for combining frame rate up-conversion and HEVC (high efficiency video coding) based on motion vector refinement |
CN110446107A (en) * | 2019-08-15 | 2019-11-12 | 电子科技大学 | A kind of video frame rate upconversion method suitable for scaling movement and light and shade variation |
CN110446107B (en) * | 2019-08-15 | 2020-06-23 | 电子科技大学 | Video frame rate up-conversion method suitable for scaling motion and brightness change |
CN111405316A (en) * | 2020-03-12 | 2020-07-10 | 北京奇艺世纪科技有限公司 | Frame insertion method, electronic device and readable storage medium |
CN112511859A (en) * | 2020-11-12 | 2021-03-16 | Oppo广东移动通信有限公司 | Video processing method, device and storage medium |
CN112954454A (en) * | 2021-02-08 | 2021-06-11 | 北京奇艺世纪科技有限公司 | Video frame generation method and device |
CN112954454B (en) * | 2021-02-08 | 2023-09-05 | 北京奇艺世纪科技有限公司 | Video frame generation method and device |
Also Published As
Publication number | Publication date |
---|---|
CN106210449B (en) | 2020-01-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106210449A (en) | The frame rate up-conversion method for estimating of a kind of Multi-information acquisition and system | |
CN104219533B (en) | A kind of bi-directional motion estimation method and up-conversion method of video frame rate and system | |
Jeong et al. | Motion-compensated frame interpolation based on multihypothesis motion estimation and texture optimization | |
CN106600536A (en) | Video imager super-resolution reconstruction method and apparatus | |
Li et al. | Multi-scale 3D scene flow from binocular stereo sequences | |
Yu et al. | Multi-level video frame interpolation: Exploiting the interaction among different levels | |
Hsu et al. | Accurate computation of optical flow by using layered motion representations | |
Seyid et al. | FPGA-based hardware implementation of real-time optical flow calculation | |
CN102263957B (en) | Search-window adaptive parallax estimation method | |
KR100987412B1 (en) | Multi-Frame Combined Video Object Matting System and Method Thereof | |
Yang et al. | Global auto-regressive depth recovery via iterative non-local filtering | |
CN103051857A (en) | Motion compensation-based 1/4 pixel precision video image deinterlacing method | |
CN106204456A (en) | Panoramic video sequences estimation is crossed the border folding searching method | |
CN104980726B (en) | A kind of binocular video solid matching method of associated movement vector | |
Irani et al. | Direct recovery of planar-parallax from multiple frames | |
CN107767393B (en) | Scene flow estimation method for mobile hardware | |
JP2004356747A (en) | Method and apparatus for matching image | |
US10432962B1 (en) | Accuracy and local smoothness of motion vector fields using motion-model fitting | |
Lin et al. | Depth map enhancement on rgb-d video captured by kinect v2 | |
CN114943911A (en) | Video object instance segmentation method based on improved hierarchical deep self-attention network | |
CN106331729A (en) | Method of adaptively compensating stereo video frame rate up conversion based on correlation | |
CN107124617A (en) | The generation method and system of random vector in motion estimation motion compensation | |
Schreer et al. | Hybrid recursive matching and segmentation-based postprocessing in real-time immersive video conferencing | |
Boltz et al. | Randomized motion estimation | |
Seo et al. | Robust 3D object tracking using an elaborate motion model |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | 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 |