CN105323420B - Method of video image processing and device - Google Patents
Method of video image processing and device Download PDFInfo
- Publication number
- CN105323420B CN105323420B CN201410368035.0A CN201410368035A CN105323420B CN 105323420 B CN105323420 B CN 105323420B CN 201410368035 A CN201410368035 A CN 201410368035A CN 105323420 B CN105323420 B CN 105323420B
- Authority
- CN
- China
- Prior art keywords
- picture frame
- pixel
- frame
- window
- kinematic parameter
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of method of video image processing and devices, belong to field of image processing.The described method includes: the N number of pixel of random acquisition in the first picture frame；Light stream tracking is carried out to N number of pixel that random sampling site obtains according to LK optical flow algorithm, obtains N number of pixel pair；According to N number of pixel to the kinematic parameter for obtaining the second picture frame；De-jitter is carried out to the second picture frame according to the kinematic parameter of second picture frame.The present invention passes through the random acquisition pixel in the first picture frame, light stream tracking is carried out to the pixel that random sampling site obtains, obtain pixel pair, according to pixel to the kinematic parameter for obtaining the second picture frame, de-jitter is carried out to the second picture frame further according to the kinematic parameter of the second picture frame, does not need to carry out feature extraction to picture frame, to enormously simplify treatment process, the processing time is reduced, achieve the effect that the real-time for improving video stabilization and saves process resource.
Description
Technical field
The present invention relates to field of image processing, in particular to a kind of method of video image processing and device.
Background technique
Video jitter, which refers in shooting process, causes video sequence there are inconsistent motion artifacts due to video camera
It shakes and fuzzy.It is compensated by extracting the true globe motion parameter of video camera, and using suitable converter technique, makes to regard
Frequency picture is smooth and stable technology is known as video stabilization technology.
The true globe motion parameter of video camera can be by carrying out image procossing acquisition to picture frame each in video.It is existing
Some video stabilization technologies extract the characteristic point in each picture frame of video to be processed first, and according to the spy extracted
Sign point carries out characteristic matching between adjacent image frame, the globe motion parameter of video camera is calculated according to matched result, finally
Original video is compensated with filtered globe motion parameter, to realize the de-jitter to video.
In the implementation of the present invention, the inventor finds that the existing technology has at least the following problems:
Existing video debounce technology needs to extract the characteristic point of each picture frame in video, and the extraction of characteristic point first
Process is complex, needs to expend the longer processing time, and the real-time of video stabilization is lower and wastes process resource.
Summary of the invention
In order to which the extraction process for solving prior art characteristic point is complex, need to expend the longer processing time, thus
The real-time of caused video stabilization it is lower and waste process resource the problem of, the embodiment of the invention provides a kind of video figures
As processing method and processing device.The technical solution is as follows:
On the one hand, a kind of method of video image processing is provided, which comprises
The N number of pixel of random acquisition in the first picture frame, the first image frame are each of composition video to be processed
Non- last frame image in picture frame, N > 3, and N are integer；
Light stream tracking is carried out to N number of pixel that random sampling site obtains according to LK optical flow algorithm, obtains N number of pixel
Right, each described pixel is to including that a pixel in N number of pixel and the pixel are corresponding second
Matched pixel point in picture frame, second picture frame are a later frame image of the first image frame；
According to N number of pixel to the kinematic parameter for obtaining second picture frame, the kinematic parameter is used to indicate
Displacement of second picture frame relative to the first image frame；
De-jitter is carried out to second picture frame according to the kinematic parameter of second picture frame.
On the other hand, a kind of video image processing device is provided, described device includes:
Pixel acquisition module, for the N number of pixel of random acquisition in the first picture frame, the first image frame is group
At the non-last frame image in each picture frame of video to be processed, N > 3, and N are integer；
Light stream tracing module, for carrying out light stream to N number of pixel that random sampling site obtains according to LK optical flow algorithm
Tracking, obtain N number of pixel pair, each described pixel to include N number of pixel in a pixel and institute
The corresponding matched pixel point in the second picture frame of pixel is stated, second picture frame is a later frame of the first image frame
Image；
Kinematic parameter obtains module, for according to N number of pixel to the kinematic parameter for obtaining second picture frame,
The kinematic parameter is used to indicate displacement of second picture frame relative to the first image frame；
Processing module, for being carried out at Key dithering according to the kinematic parameter of second picture frame to second picture frame
Reason.
Technical solution provided in an embodiment of the present invention has the benefit that
By the random acquisition pixel in the first picture frame, light stream is carried out to N number of pixel that random sampling site obtains and is chased after
Track obtains N number of pixel pair, according to N number of pixel to the kinematic parameter for obtaining the second picture frame, according to the fortune of the second picture frame
Dynamic parameter carries out de-jitter to the second picture frame, the pixel of random acquisition can be determined adjacent by LK optical flow algorithm
Matched pixel point in picture frame then determines the kinematic parameter of adjacent image frame, does not need to carry out feature extraction to picture frame,
To enormously simplify treatment process, the processing time is reduced, reach the real-time for improving video stabilization and saves process resource
Effect.
Detailed description of the invention
To describe the technical solutions in the embodiments of the present invention more clearly, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, drawings in the following description are only some embodiments of the invention, for
For those of ordinary skill in the art, without creative efforts, it can also be obtained according to these attached drawings other
Attached drawing.
Fig. 1 is the method flow diagram of method of video image processing provided by one embodiment of the present invention；
Fig. 2 be another embodiment of the present invention provides method of video image processing method flow diagram；
Fig. 3 be another embodiment of the present invention provides pixel acquisition window be arranged schematic diagram；
Fig. 4 be another embodiment of the present invention provides video stabilization schematic diagram；
Fig. 5 is the structure drawing of device of video image processing device provided by one embodiment of the present invention；
Fig. 6 be another embodiment of the present invention provides video image processing device structure drawing of device；
Fig. 7 is the structural block diagram of electronic equipment provided by one embodiment of the present invention.
Specific embodiment
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with attached drawing to embodiment party of the present invention
Formula is described in further detail.
Referring to FIG. 1, it illustrates the method flow diagrams of method of video image processing provided by one embodiment of the present invention.
The method of video image processing can be used for video processing equipment, for example smart phone, video camera, tablet computer, e-book are read
It reads in the equipment such as device and PC, which may include:
Step 102, the N number of pixel of random acquisition in the first picture frame, the first picture frame are composition video to be processed
Non- last frame image in each picture frame, N > 3, and N are integer；
Step 104, light stream tracking is carried out to N number of pixel that random sampling site obtains according to LK optical flow algorithm, obtains N number of picture
Vegetarian refreshments pair, each pixel is to including that a pixel in N number of pixel and the pixel are corresponding in the second picture frame
In matched pixel point, the second picture frame be the first picture frame a later frame image；
Wherein, light stream is the instantaneous velocity of pixel motion of the space motion object on observation imaging surface.LK(Lucas-
Kanade) optical flow algorithm is a kind of light stream evaluation method of two frame differences, can be used for estimating each pixel in video image
In t moment to the movement of the position between t+ time Δt.
Step 106, according to N number of pixel to the kinematic parameter for obtaining the second picture frame, kinematic parameter is used to indicate second
Displacement of the picture frame relative to the first picture frame；
Step 108, de-jitter is carried out to the second picture frame according to the kinematic parameter of the second picture frame.
In conclusion method of video image processing provided in an embodiment of the present invention, by being adopted at random in the first picture frame
Collect pixel, light stream tracking is carried out to N number of pixel that random sampling site obtains, N number of pixel pair is obtained, according to N number of pixel
To the kinematic parameter for obtaining the second picture frame, the second picture frame is carried out at Key dithering according to the kinematic parameter of the second picture frame
Reason, can determine matched pixel point of the pixel of random acquisition in adjacent image frame by LK optical flow algorithm, then determine
The kinematic parameter of adjacent image frame does not need to carry out feature extraction to picture frame, so that treatment process is enormously simplified, at reduction
The time is managed, achieve the effect that the real-time for improving video stabilization and saves process resource.
Referring to FIG. 2, it illustrates another embodiment of the present invention provides method of video image processing method flow diagram.
The method of video image processing can be used in video processing equipment, such as smart phone, video camera, tablet computer, e-book
In the equipment such as reader and PC, de-jitter is carried out to video to be processed.The method of video image processing can
To include:
Step 202, pixel acquisition window is determined in the first picture frame；
The pixel acquisition window is in the inside of first picture frame, and the boundary of the pixel acquisition window and this
The boundary of one picture frame is not overlapped.
Since the parts of images content that video jitter may cause former frame removes image boundary in next frame, if entire
Random sampling site in picture frame, the collected pixel of boundary for being likely to appear in picture frame are not present in next frame image
Situation, such point is referred to as " bad point ", thus during carrying out subsequent operation, the presence of bad point is easy influence video and goes
The effect of shake.Therefore, before random acquisition pixel, an acquisition window can be defined first in picture frame as picture
The boundary of vegetarian refreshments acquisition, keeps appropriately distance, only in image between the boundary of the Image Acquisition window and the boundary of picture frame
The stochastical sampling of pixel is carried out in pixel acquisition window in frame, to reduce the probability for collecting bad point, improves video
The effect of Key dithering.
Preferably, in method provided in this embodiment, when determining pixel acquisition window, light stream tracking can also be obtained
Tracking window；It keeps the center of the tracking window constant, sets k times of primary side circle for the boundary of the tracking window,
Determine that the window obtained after setting is the pixel acquisition window, k is positive number.
Wherein, the tracking window of light stream tracking is the pre-set window of developer, for subsequent to collected
Pixel carries out light stream tracking.It, can be in order to further increase the accuracy rate of light stream tracking in method provided in this embodiment
According to the tracking window of light stream tracking, pixel acquisition window is set.For example, the setting of pixel acquisition window can be such as Fig. 3 institute
Show, wherein the tracking window 31 of light stream tracking is that developer is pre-set in picture frame 30, in setting pixel collecting window
It when mouth, keeps the center of tracking window 31 constant, the frontier distance of window will be tracked according to scheduled ratio according to actual needs
It zooms in or out k times, obtained new window 32, the new window 32 may act as pixel acquisition window, typical
, in the present embodiment, can take the value of k is 1.5, i.e., is extended to the up-and-down boundary for tracking window and right boundary original
1.5 times, the new window of acquisition is as pixel acquisition window.
Step 204, the N number of pixel of random acquisition, N > 3, and N are integer in the pixel acquisition window；
Wherein, the first picture frame is the non-last frame image in each picture frame for form video to be processed.
In random acquisition pixel point, the pixel in pixel acquisition window can be generated by random number generator
Coordinate.Preferably, in order to which optimal Key dithering effect can be reached while simplifying the complexity of calculating process as far as possible,
In the present embodiment, it is 50, i.e. 50 pixels of random acquisition that N, which can be set,.
Step 206, light stream tracking is carried out to N number of pixel that random sampling site obtains according to LK optical flow algorithm, obtains N number of picture
Vegetarian refreshments pair；
Wherein, each pixel is to including that a pixel in N number of pixel and the pixel are corresponding second
Matched pixel point in picture frame, the second picture frame are a later frame image of the first picture frame.
Wherein, light stream is the instantaneous velocity of pixel motion of the space motion object on observation imaging surface.LK optical flow algorithm
It is a kind of light stream evaluation method of two frame differences, can be used for estimating the pixel in video image by t moment to t+ time Δt
Position movement.
When carrying out pixel point tracking according to LK optical flow algorithm, it is with N number of pixel collected from the first picture frame
Input, is tracked by light stream, obtains matched pixel point of this N number of pixel respectively in the second picture frame.The step 206 can be with
Including following sub-step:
1) the corresponding multi-layer image pyramid of the second picture frame of construction；
In the image pyramid, the image of higher level is the smoothed out down-sampling form of lower image, the second picture frame
The original image number of plies be 0 layer, i.e. the bottom.
2) light stream tracking is successively carried out by high-rise image to low layer pictures, obtains the second picture frame relative to the first picture frame
Motion vector；
After original image is smoothly down-sampled to certain level, the motion vector of adjacent interframe pixel will become enough
It is small, meet the constraint condition of optical flow computation, can directly carry out light stream estimation.When actually calculating, by high-rise image to low layer
Image successively carries out light stream tracking, after the light stream incremental computations of the image of a certain level come out, will calculate the light stream obtained and increases
Amount is added on the optical flow computation initial value of the tomographic image, the optical flow computation initial value as its next tomographic image, wherein highest tomographic image
Optical flow computation initial value can be set to 0 or other values.This process constantly carries out, until calculating in pyramid most bottom
The light stream increment of the original image of second picture frame of layer, by the light stream meter of the light stream increment of the original image and the original image
The sum of initial value is calculated as the first picture frame to the light stream of the second picture frame, which is the second picture frame relative to the first image
The motion vector of frame.
3) basis collected N number of pixel and motion vector from the first picture frame obtains N number of pixel pair
It should be in the matched pixel point in the second picture frame.
The coordinate of N number of pixel collected from the first picture frame is added into the motion vector, it can it is N number of to obtain this
Coordinate of the corresponding matched pixel o'clock of pixel in the second picture frame.
Wherein, above-mentioned steps need to pre-set the size of the pyramid number of plies and light stream tracking window.
In the present embodiment, the needs of light stream tracking can be met by being 3 layers for the pyramid number of plies.Meanwhile in order to make
The movement for being capable of handling biggish image-region is tracked in light stream, and a biggish tracking window can be set, for example, setting tracking
Window size is 20 pixels × 20 pixels.
Step 208, according to N number of pixel to the kinematic parameter for obtaining the second picture frame, kinematic parameter is used to indicate the
Displacement of two picture frames relative to the first picture frame；
In method shown in the present embodiment, it can estimate that the movement of the second picture frame is joined by random sampling unification algorism
Number.Specifically, when obtaining the kinematic parameter of the second picture frame, available motion model compatible with the kinematic parameter,
The motion model is affine model, and determines the frequency in sampling K of random sampling unification algorism；Then according to the affine model and this
Frequency in sampling K carries out the calculating of random sampling unification algorism, extracts n pixel pair from N number of pixel centering, wherein 3≤n
< N；Further according to the n pixel of extraction to the model parameter for calculating the affine model with the affine model, then root
According to the kinematic parameter of the second picture frame of the affine model and model parameter calculation.
Wherein, when the n pixel according to extraction is to the model parameter for calculating the affine model with the affine model,
Second image can be calculated by singular value decomposition method or LU factorization by the n pixel to the affine model is inputted
The kinematic parameter of frame.
In the method, select affine model for global motion model.Affine model is that a kind of two-dimensional coordinate is sat to two dimension
Linear transformation between mark keeps the grazing (straight line is still straight line after converting) and collimation (i.e. X-Y scheme of X-Y scheme
Relative positional relationship between shape is constant, is still parallel lines by transformation parallel lines).Affine transformation can pass through a series of atom
Convert compound realization, comprising: translation scales, and overturning, rotation and mistake are cut.Due to video jitter it is most important the reason is that rotation and
Translation, and affine transformation can be used to indicate that this type games.Affine model can use the transformation matrix expression of 3 × 3 sizes,
Its front two row element is unknown parameter, and last line is (0,0,1), and former coordinate (x, y) is transformed to new coordinate by transformation matrix
(x ', y '), former coordinate and new coordinate are all considered as the three of the last behavior 1 and rank vector here.In construction Solving Linear
During, it is also necessary to carry out such as down conversion:
Solving Linear be solve unknown vector in the case where knowing equation group coefficient matrix and right side vector, and
We need affine model parameter to be requested to be equivalent to equation group coefficient herein, and former coordinate is equivalent to unknown vector, by by mould
The position of shape parameter and former coordinate, which is exchanged, can be used the method solving model parameter for solving system of linear equations, mapping mode
It is as follows:
It is transformed to
Wherein a~f is affine model parameter, and (x, y) is former coordinate, (x ', y ') it is new coordinate.
The N pixel obtained in above-mentioned steps 206 by LK optical flow algorithm can imitate input affine model
Penetrate the parameter Estimation of model.However, most point is identical to having in the matching result obtained by LK optical flow algorithm
Movement tendency, but there is also the motion vector of some points pair exceptions, that is, pass through the matched pixel point and reality of light stream tracking acquisition
Matched pixel point between error it is larger, these abnormal points are to being properly termed as " point not in the know to ", the lesser point pair of remaining error
" intra-office point to " is properly termed as if the number of point pair not in the know is more or deviation is larger, the subsequent knot for calculating kinematic parameter will be made
Fruit error.Random sampling unification algorism RANSAC (Random Sample Consensus) plays a game exterior point to there is good robust
Property, even if the number of point pair not in the know, close to half, RANSAC is still it can be concluded that correct result.It is provided in an embodiment of the present invention
Method is extracted by random sampling unification algorism to N pixel to multiple random sampling is carried out through light stream tracking acquisition
The lesser n pixel pair of error, specific steps can be such that between match point and actual match point
1) an alternative point is extracted at random to collection from N number of pixel centering, the alternative point is a to the n comprising randomly selecting is concentrated
Pixel pair, according to the pixel extracted to the model parameter for calculating affine model；
2) according to the model parameter and affine model being calculated to the rest of pixels point pair of N number of matched pixel point centering
It is tested, calculates the error of remaining each pixel pair.
Specifically, step 1 is calculated to the model parameter obtained substitutes into affine model, affine matrix is obtained；It is affine by this
For matrix to rest of pixels point centering, the coordinate of the pixel extracted from the first picture frame carries out affine transformation, obtains the picture
The coordinate of the corresponding new point of vegetarian refreshments；The error between the coordinate of the new point and the coordinate of the matched pixel point of the pixel is calculated,
The error is the error for the pixel pair that this test obtains.
3) pixel that error is less than error threshold counts the number of the intra-office point pair to intra-office point pair is determined as；
4) above-mentioned 1~3 step is repeated K times, obtains K alternative points to collection；Determine most alternative of corresponding intra-office point logarithm
Point is to collection；
5) judge whether the determining alternative point is big to the quantitative proportion in the N pixel pair to corresponding intra-office point is collected
In proportion threshold value, if so, by the determining alternative point to n pixel of concentration to being retrieved as random sampling unification algorism
Extract result.
Wherein, above-mentioned random sampling unification algorism needs developer that following parameter: random sampling points n, error is arranged
The accuracy z of threshold value, proportion threshold value and random sampling unification algorism result.
Random sampling points n depending on model parameter, since affine model parameter is 6, is at least needed under normal conditions
Want 3 points to can just solve a model parameter, therefore, random sampling points take n value to be at least 3, in the present embodiment
In, the value of n is specifically as follows 3~6.
Error threshold is to judge a point to whether being classified as the threshold value of intra-office point pair；In the present embodiment, should
Error threshold can be set to 0.5.
Proportion threshold value is for judging the whether accurate threshold value of the result of this random sampling unification algorism；If K calculating
The ratio that maximum value in the intra-office point logarithm of acquisition accounts for the N pixel logarithm is less than the proportion threshold value, then illustrates that this is random
Unification algorism result of sampling inaccuracy, in the present embodiment, which can be set to 0.5.
The accuracy z of random sampling unification algorism result is used to estimate the frequency in sampling K of random sampling unification algorism, at this
In embodiment, z can be set to 0.99.Frequency in sampling K is calculated according to following empirical equation:
K=log (1-z)/log (1-wm)；
Wherein w is point in collection (the N number of pixel obtained by LK optical flow algorithm the to) point taken to falling in
Probability in error range, it is assumed that it puts to being concentrated with the point of half to falling in except error range, it may be considered that w=0.5, this
In embodiment, w=0.4 can be taken；M is to solve for model parameter and at least needs a little to the number of the element of concentration, above-mentioned affine mould
There are 6 parameters to need to solve in type, and available two equations of each pair of point, therefore, it is necessary to m >=3 are arranged；Z refers to will be upper
It states process to repeat K times, for the n point randomly selected at least once to the probability of all intra-office points pair, i.e. arithmetic result is correct
Probability, value z=0.99.
According to above-mentioned empirical equation and parameter value, it can be deduced that above-mentioned sampling process is repeated 19 times by K=19, calculate
Method show that the probability of correct result is 0.99.
When according to the kinematic parameter of the second picture frame of affine model and model parameter calculation, consistent according to random sampling
Algorithm extracts n pixel to rear, can be according to the determining target point to collecting corresponding each intra-office point to recalculating
The model parameter for recalculating acquisition is substituted into affine model, obtains new affine matrix, then root by the model parameter of affine model
The kinematic parameter of second picture frame is calculated according to the affine matrix.
Step 210, determine Gaussian smoothing window, include in the Gaussian smoothing window second picture frame and this second
Each M picture frame before and after picture frame, M is positive integer；
Gaussian smoothing carries out in the smooth window of setting, and for smooth window centered on to smooth picture frame, front and back is each
There is M picture frame, i.e., altogether includes 2M+1 picture frame in smooth window.Under normal conditions, the value range of M can be set
Between 10~30.
Step 212, according to the kinematic parameter of each picture frame in the Gaussian smoothing window to the fortune of second picture frame
It is smooth that dynamic parameter carries out local Gaussian；
Specifically, can be according to the beginning parameter transform model of each picture frame in the Gaussian smoothing window second picture frame
Each m picture frame in front and back respectively correspond tos the relative movement parameters of second picture frame；Centered on second picture frame, to this
The relative movement parameters that each m picture frame respectively correspond tos second picture frame before and after second picture frame carry out Gaussian smoothing.
It is 41 that this, which sentences smooth window size, comprising for 20 frame each before and after smoothed frame, when after smoothed frame after 20
After the action reference variable of frame, each frame is to the relative movement parameters to smoothed frame in calculation window.Calculate relative motion
Method is continuously to be transported using the kinematic parameter of the adjacent interframe calculated by a certain frame and to the N between smoothed frame
Dynamic parameter is multiplied, and obtained kinematic parameter is the frame to the relative movement parameters to smoothed frame, such as: it is to flat with the 10th frame
Sliding frame, the relative movement parameters of the 8th frame to the 10th frame are the kinematic parameter of the 8th frame to the 9th frame multiplied by the fortune of the 9th frame to the 10th frame
Dynamic parameter；The relative movement parameters of 12nd frame to the 10th frame are the kinematic parameter of the 10th frame to the 11st frame multiplied by the 11st frame to the 12nd
The kinematic parameter of frame.Next in window relative to smoothed frame each relative movement parameters carry out Gaussian smoothing, into
When row Gaussian smoothing, smoothing factor is assigned for each picture frame in smooth window first, according to the smoothing factor of imparting to each
The corresponding relative movement parameters of a picture frame carry out Gaussian smoothing, obtain the weighted average of each relative movement parameters, should plus
Weight average value is the smoothed out kinematic parameter to smoothed frame.It should be noted that each smoothing factor is that developer is pre-
The parameter being first arranged, and the corresponding smoothing factor of each picture frame meets Gaussian Profile and normalization centered on to smoothed frame
Principle.Wherein, meeting Gaussian Profile is determined by the correlation of adjacent interframe, since the picture frame of composition video is closer,
Its correlation is bigger, i.e., to the maximum of gaussian coefficient corresponding to smoothed frame, the corresponding smooth system of outermost two frame of smooth window
Number is minimum.The Gaussian smoothing window is translated with the processing of video frame, as long as the fortune for having video frame to meet its 20 frame of front and back
Dynamic parameter is all estimated to complete, and can carry out to it smoothly, smooth rear hatch center moves to next frame.In the present embodiment, it carries out
Local Gaussian can smoothly make the video after debounce have continuity, to achieve the effect that removal shake.
Step 214, second picture frame is carried out at Key dithering according to the kinematic parameter of smoothed out second picture frame
Reason.
Wherein, to the second picture frame carry out de-jitter when, can according to the smoothed out kinematic parameter to this second
Parameters in picture frame carry out image and redraw, second picture frame that obtains that treated.
When restoring to shake video, according to the processing method in above-mentioned steps, to each of composition video to be processed
A picture frame carries out de-jitter respectively, it can the video after obtaining Key dithering.
It should be noted that can be filled out at this time using boundary since affine transformation may make parts of images remove boundary
Reinforcing method carries out boundary compensation.
Method provided in this embodiment is detected instead of characteristic point using random sampling site, reduces calculation amount, so that algorithm has
There is higher real-time, meanwhile, local smoothing method is carried out to kinematic parameter using Gaussian smoothing, the kinematic parameter knot that estimation is obtained
The kinematic parameter for closing consecutive frame carries out that smoothly, the continuity between image frame adjacent in video can be improved, thus further
Improve the effect of removal shake.
Below with reference to specific attached drawing and example, the present embodiment above method is described in detail.
The process for the removal video jitter method that this example provides is as shown in Figure 4: by taking the video of yuv format as an example, first
The original video files of yuv format are read, create first frame, and read next frame.After reading next frame, to present frame
And its former frame carries out kinematic parameter estimation.In this example, setting Gaussian smoothing window size is 41 frames, comprising to smoothed frame
Preceding 20 frame and rear 20 frame, due to not having video frame and corresponding kinematic parameter before first frame, in order to meet local smoothing method
Versatility, polishing, polishing method can be carried out to the kinematic parameter before first frame are as follows: estimation shown in Fig. 4 movement is joined
After several file first frame images, setting is unit battle array for the kinematic parameter of the preceding 20 frame image of polishing.According to window
Parameter setting, Gaussian smoothing also need 20 frame after smoothed frame, can be with finally when the frame number after smoothed frame is less than 20 frame
The kinematic parameter of one frame carries out polishing.Specifically, during estimating kinematic parameter, when the movement ginseng of a frame image is completed in estimation
When counting and reading next frame, judge whether next frame reads success, successfully estimates former frame to the present frame read if reading
Kinematic parameter, if reading unsuccessful, it is determined that the frame image for estimating completion before is last frame image, and it is last that this is arranged
The kinematic parameter of one frame image is the kinematic parameter of the rear 20 frame image for polishing.
The estimation of kinematic parameter is successively carried out since the first frame of original video, when the 21st frame image in original video
After the completion of kinematic parameter estimation, that is, start to carry out the smooth of kinematic parameter since first frame image, as long as specifically, original view
The action reference variable of picture frame of a certain frame number greater than 20 is completed in frequency, it can by the 20th frame image before the picture frame
It is set as calculating the relative movement parameters for waiting for smoothed frame to this to 20 frame each before and after smoothed frame to smoothed frame, and to be to smoothed frame
Center, to assign smoothing factor to smoothed frame and to 20 frame each before and after smoothed frame, and with the smoothing factor assigned to smoothing windows
The corresponding relative movement parameters of each frame image in mouthful carry out Gaussian smoothing, obtain the weighted average of each relative movement parameters
Value, which is the smoothed out kinematic parameter to smoothed frame, after the completion of the smooth of motion paramter to a certain frame,
The center of smooth window can move to next frame, until carrying out to the kinematic parameter of the last frame image in original video flat
Sliding and Key dithering terminates, the video after obtaining Key dithering；It should be noted that being put down centered on the kinematic parameter to smoothed frame
Sliding coefficient meets Gaussian Profile and normalization principle.
In addition, can be divided into random sampling site, light stream tracking and random sampling consistent for the estimation steps of above-mentioned kinematic parameter
Algorithm.
In random sampling site, the part in the pixel acquisition window boundary in previous frame image can be adopted at random
The coordinate of point, the pixel of acquisition is generated by random number generator.In this example, the size of acquisition window can be defined as
1.5 times of the light stream tracking window size being arranged in LK optical flow algorithm, i.e. the right boundary of pixel acquisition window is defined as light
1.5 times of stream tracking window width, up-and-down boundary are defined as 1.5 times of light stream tracking window height.In this example, it adopts at random
The pixel quantity of collection is 50, can be basic when the value range for adopting points N at random is 30 to 50 by experimental analysis
Meet functional requirement.
When carrying out light stream tracking, it is 3 layers that the pyramid number of plies, which can be set, and window size is that 20 pixels are wide, and 20 pixels are high,
By LK optical flow algorithm, the determining matched pixel of pixel adopted at random with former frame, obtains 50 in latter consecutive frame
A matched pixel point pair.
Finally, extracting suitable office from matched pixel point centering by random sampling unification algorism and preset affine model
Interior point pair, and kinematic parameter is estimated according to result is extracted.In this example, rule of thumb can to calculate random sampling consistent for formula
The frequency in sampling of algorithm is 19 times, when each random sampling, and reasonable random sampling points range is 3 to 6 points.
In conclusion method of video image processing provided in an embodiment of the present invention, by being adopted at random in the first picture frame
Collect pixel, light stream tracking is carried out to N number of pixel that random sampling site obtains, N number of pixel pair is obtained, according to N number of pixel
To the kinematic parameter for obtaining the second picture frame, the second picture frame is carried out at Key dithering according to the kinematic parameter of the second picture frame
Reason, can determine matched pixel point of the pixel of random acquisition in adjacent image frame by LK optical flow algorithm, then determine
The kinematic parameter of adjacent image frame does not need to carry out feature extraction to picture frame, so that treatment process is enormously simplified, at reduction
The time is managed, achieve the effect that the real-time for improving video stabilization and saves process resource.
In addition, method of video image processing provided in an embodiment of the present invention, by the way that pixel acquisition window is arranged, in pixel
Random acquisition pixel in the boundary of point acquisition window, reduces and collects the probability of bad point, improve video stabilization efficiency and
Accuracy.
In addition, method of video image processing provided in an embodiment of the present invention, true by the tracking window tracked according to light stream
Determine pixel acquisition window, further increases the treatment effeciency of subsequent light stream tracking.
Referring to FIG. 5, it illustrates the structure drawing of device of video image processing device provided by one embodiment of the present invention.
The video image processing device can be used for executing method of video image processing as shown in Figure 1 or 2, the video image processing
Device may include:
Pixel acquisition module 301, for the N number of pixel of random acquisition, the first image frame in the first picture frame
The non-last frame image in each picture frame to form video to be processed, N > 3 and N are integer；
Light stream tracing module 302, for carrying out light to N number of pixel that random sampling site obtains according to LK optical flow algorithm
Stream tracking, obtain N number of pixel pair, each described pixel to include N number of pixel in a pixel and
The corresponding matched pixel point in the second picture frame of the pixel, second picture frame are the latter of the first image frame
Frame image；
Kinematic parameter obtains module 303, for being joined according to N number of pixel to the movement for obtaining second picture frame
Number, the kinematic parameter are used to indicate displacement of second picture frame relative to the first image frame；
Processing module 304, for carrying out debounce to second picture frame according to the kinematic parameter of second picture frame
Dynamic processing.
In conclusion video image processing device provided in an embodiment of the present invention, by being adopted at random in the first picture frame
Collect pixel, light stream tracking is carried out to N number of pixel that random sampling site obtains, N number of pixel pair is obtained, according to N number of pixel
To the kinematic parameter for obtaining the second picture frame, the second picture frame is carried out at Key dithering according to the kinematic parameter of the second picture frame
Reason, can determine matched pixel point of the pixel of random acquisition in adjacent image frame by LK optical flow algorithm, then determine
The kinematic parameter of adjacent image frame does not need to carry out feature extraction to picture frame, so that treatment process is enormously simplified, at reduction
The time is managed, achieve the effect that the real-time for improving video stabilization and saves process resource.
Referring to FIG. 6, it illustrates another embodiment of the present invention provides video image processing device structure drawing of device.
The video image processing device can be used for executing method of video image processing as shown in Figure 1 or 2, the video image processing
Device can be implemented as the electronic equipments such as smart phone, video camera, tablet computer, E-book reader and PC or
It is a part of in electronic equipment.The video image processing device may include:
Pixel acquisition module 401, for the N number of pixel of random acquisition, the first image frame in the first picture frame
The non-last frame image in each picture frame to form video to be processed, N > 3 and N are integer；
Light stream tracing module 402, for carrying out light to N number of pixel that random sampling site obtains according to LK optical flow algorithm
Stream tracking, obtain N number of pixel pair, each described pixel to include N number of pixel in a pixel and
The corresponding matched pixel point in the second picture frame of the pixel, second picture frame are the latter of the first image frame
Frame image；
Kinematic parameter obtains module 403, for being joined according to N number of pixel to the movement for obtaining second picture frame
Number, the kinematic parameter are used to indicate displacement of second picture frame relative to the first image frame；
Processing module 404, for carrying out debounce to second picture frame according to the kinematic parameter of second picture frame
Dynamic processing.
The pixel acquisition module 401, comprising:
Acquisition window determination unit 4011, for determining that pixel acquisition window, the pixel acquisition window are in institute
The inside of the first picture frame is stated, and the boundary of the pixel acquisition window is not overlapped with the boundary of the first image frame；
Acquisition unit 4012, for the random acquisition pixel in the pixel acquisition window.
The window determination unit 4011, comprising:
Window obtains subelement 4011a, for obtaining the tracking window of the light stream tracking；
Window determines subelement 4011b, for keeping the center of the tracking window constant, by the tracking window
Boundary be set as k times of primary side circle, determine that the window obtained after setting is the pixel acquisition window, k is positive number.
The kinematic parameter obtains module 403, comprising:
Model acquiring unit 4031, for obtaining motion model compatible with the kinematic parameter, the motion model
For affine model；
Frequency in sampling determination unit 4032, for determining the frequency in sampling K of random sampling unification algorism；
Extraction unit 4033, for carrying out random sampling unification algorism according to the affine model and the frequency in sampling K
It calculates, extracts n pixel pair, 3≤n < N from N number of pixel centering；
Computing unit 4034, for calculating described second to the affine model according to the n pixel of extraction
The kinematic parameter of picture frame.
The computing unit 4034, for the n pixel to the affine model is inputted, to be passed through singular value decomposition
Method or LU factorization calculate the model parameter of the affine model, according to the affine model and model parameter calculation institute
State the kinematic parameter of the second picture frame.
The processing module 404, comprising:
Smooth window determination unit 4041 includes in the Gaussian smoothing window for determining Gaussian smoothing window
Each M picture frame of the second picture frame and second picture frame front and back is stated, M is positive integer；
Smooth unit 4042, for the kinematic parameter according to each picture frame in the Gaussian smoothing window to described
It is smooth that the kinematic parameter of two picture frames carries out local Gaussian；
Processing unit 4043, for according to the kinematic parameter of smoothed out second picture frame to second picture frame
Carry out de-jitter.
In conclusion video image processing device provided in an embodiment of the present invention, by being adopted at random in the first picture frame
Collect pixel, light stream tracking is carried out to N number of pixel that random sampling site obtains, N number of pixel pair is obtained, according to N number of pixel
To the kinematic parameter for obtaining the second picture frame, the second picture frame is carried out at Key dithering according to the kinematic parameter of the second picture frame
Reason, can determine matched pixel point of the pixel of random acquisition in adjacent image frame by LK optical flow algorithm, then determine
The kinematic parameter of adjacent image frame does not need to carry out feature extraction to picture frame, so that treatment process is enormously simplified, at reduction
The time is managed, achieve the effect that the real-time for improving video stabilization and saves process resource.
In addition, video image processing device provided in an embodiment of the present invention, by the way that pixel acquisition window is arranged, in pixel
Random acquisition pixel in the boundary of point acquisition window, reduces and collects the probability of bad point, improve video stabilization efficiency and
Accuracy.
In addition, video image processing device provided in an embodiment of the present invention, true by the tracking window tracked according to light stream
Determine pixel acquisition window, further increases the treatment effeciency of subsequent light stream tracking.
Referring to FIG. 7, the structural block diagram of the electronic equipment provided it illustrates an embodiment of the present disclosure.The electronics
Equipment can be smart phone, video camera, tablet computer, E-book reader and PC etc. with certain processing capacity
Smart machine.The electronic equipment 500 includes central processing unit (CPU) 501 including random access memory (RAM) 502
With the system storage 504 of read-only memory (ROM) 503, and connection system storage 504 and central processing unit 501
System bus 505.The electronic equipment 500 further includes that the substantially defeated of information is transmitted between each device helped in computer
Enter/output system (I/O system) 506, and for storage program area 513, application program 512 and other program modules 515
Mass-memory unit 507.
The basic input/output 506 includes display 508 for showing information and inputs letter for user
The input equipment 509 of such as mouse, keyboard etc of breath.Wherein the display 508 and input equipment 509 are all by being connected to
The input and output controller 510 of system bus 505 is connected to central processing unit 501.The basic input/output 506
Can also include input and output controller 510 with for receive and handle from keyboard, mouse or electronic touch pen etc. it is multiple its
The input of his equipment.Similarly, input and output controller 510 also provides output to display screen, printer or other kinds of defeated
Equipment out.
The mass-memory unit 507 is by being connected to the bulk memory controller (not shown) of system bus 505
It is connected to central processing unit 501.The mass-memory unit 507 and its associated computer-readable medium set for electronics
Standby 500 provide non-volatile memories.That is, the mass-memory unit 507 may include such as hard disk or CD-
The computer-readable medium (not shown) of ROM drive etc.
Without loss of generality, the computer-readable medium may include computer storage media and communication media.Computer
Storage medium includes information such as computer readable instructions, data structure, program module or other data for storage
The volatile and non-volatile of any method or technique realization, removable and irremovable medium.Computer storage medium includes
RAM, ROM, EPROM, EEPROM, flash memory or other solid-state storages its technologies, CD-ROM, DVD or other optical storages, tape
Box, tape, disk storage or other magnetic storage devices.Certainly, skilled person will appreciate that the computer storage medium
It is not limited to above-mentioned several.Above-mentioned system storage 504 and mass-memory unit 507 may be collectively referred to as memory.
According to various embodiments of the present invention, the electronic equipment 500 can also be connected to the network by internet etc.
Remote computer operation on to network.Namely electronic equipment 500 can be by the network that is connected on the system bus 505
Interface unit 511 is connected to network 512, in other words, Network Interface Unit 511 can be used also to be connected to other kinds of net
Network or remote computer system (not shown).
The memory further includes that one or more than one program, the one or more programs are stored in
In memory, the one or more programs have following function:
The N number of pixel of random acquisition in the first picture frame, the first image frame are each of composition video to be processed
Non- last frame image in picture frame, N > 3, and N are integer；
Light stream tracking is carried out to N number of pixel that random sampling site obtains according to LK optical flow algorithm, obtains N number of pixel
Right, each described pixel is to including that a pixel in N number of pixel and the pixel are corresponding second
Matched pixel point in picture frame, second picture frame are a later frame image of the first image frame；
According to N number of pixel to the kinematic parameter for obtaining second picture frame, the kinematic parameter is used to indicate
Displacement of second picture frame relative to the first image frame；
De-jitter is carried out to second picture frame according to the kinematic parameter of second picture frame.
Wherein, the N number of pixel of random acquisition in the first picture frame, comprising:
Determine that pixel acquisition window, the pixel acquisition window are in the inside of the first image frame, and described
The boundary of pixel acquisition window is not overlapped with the boundary of the first image frame；
The random acquisition pixel in the pixel acquisition window.
The determining pixel acquisition window, comprising:
Obtain the tracking window of the light stream tracking；
It keeps the center of the tracking window constant, sets k times of primary side circle for the boundary of the tracking window,
Determine that the window obtained after setting is the pixel acquisition window, k is positive number.
It is described according to N number of pixel to the kinematic parameter for obtaining second picture frame, comprising:
Motion model compatible with the kinematic parameter is obtained, the motion model is affine model；
Determine the frequency in sampling K of random sampling unification algorism；
The calculating of random sampling unification algorism is carried out according to the affine model and the frequency in sampling K, from N number of pixel
N pixel pair, 3≤n < N are extracted in point centering；
According to the n pixel of extraction to the kinematic parameter for calculating second picture frame with the affine model.
The n pixel according to extraction joins the movement that second picture frame is calculated with the affine model
Number, comprising:
By the n pixel to the affine model is inputted, institute is calculated by singular value decomposition method or LU factorization
State the model parameter of affine model；
According to the kinematic parameter of the second picture frame described in the affine model and the model parameter calculation.
The kinematic parameter according to second picture frame carries out de-jitter to second picture frame, comprising:
It determines Gaussian smoothing window, includes second picture frame and second figure in the Gaussian smoothing window
As M picture frame each before and after frame, M is positive integer；
The movement of second picture frame is joined according to the kinematic parameter of each picture frame in the Gaussian smoothing window
It is smooth that number carries out local Gaussian；
De-jitter is carried out to second picture frame according to the kinematic parameter of smoothed out second picture frame.
In conclusion electronic equipment provided in an embodiment of the present invention, by the random acquisition pixel in the first picture frame,
Light stream tracking is carried out to N number of pixel that random sampling site obtains, N number of pixel pair is obtained, according to N number of pixel to acquisition second
The kinematic parameter of picture frame carries out de-jitter to the second picture frame according to the kinematic parameter of the second picture frame, passes through LK light
Flow algorithm can determine matched pixel point of the pixel of random acquisition in adjacent image frame, then determine adjacent image frame
Kinematic parameter does not need to carry out feature extraction to picture frame, to enormously simplify treatment process, reduces the processing time, reaches
It improves the real-time of video stabilization and saves the effect of process resource.
In addition, electronic equipment provided in an embodiment of the present invention, by the way that pixel acquisition window is arranged, in pixel collecting window
Random acquisition pixel in the boundary of mouth reduces the probability for collecting bad point, improves the efficiency and accuracy of video stabilization.
In addition, electronic equipment provided in an embodiment of the present invention, determines pixel by the tracking window tracked according to light stream
Acquisition window further increases the treatment effeciency of subsequent light stream tracking.
It should be understood that video image processing device provided by the above embodiment is when handling video image, only more than
The division progress of each functional module is stated for example, can according to need and in practical application by above-mentioned function distribution by difference
Functional module complete, i.e., the internal structure of device is divided into different functional modules, with complete it is described above whole or
Person's partial function.In addition, video image processing device provided by the above embodiment belongs to method of video image processing embodiment
Same design, specific implementation process are detailed in embodiment of the method, and which is not described herein again.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Those of ordinary skill in the art will appreciate that realizing that all or part of the steps of above-described embodiment can pass through hardware
It completes, relevant hardware can also be instructed to complete by program, the program can store in a kind of computer-readable
In storage medium, storage medium mentioned above can be read-only memory, disk or CD etc..
The foregoing is merely presently preferred embodiments of the present invention, is not intended to limit the invention, it is all in spirit of the invention and
Within principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (7)
1. a kind of method of video image processing, which is characterized in that the described method includes:
Obtain the tracking window of light stream tracking；
It keeps the center of the tracking window constant, sets k times of primary side circle for the boundary of the tracking window, determine
The window obtained after setting is pixel acquisition window, and the pixel acquisition window is in the inside of the first picture frame, and institute
The boundary for stating pixel acquisition window is not overlapped with the boundary of the first image frame, and the first image frame is that composition is to be processed
Non- last frame image in each picture frame of video, k is positive number；
The N number of pixel of random acquisition, N > 3, and N are integer in the pixel acquisition window；
Light stream tracking is carried out to N number of pixel that random sampling site obtains according to LK optical flow algorithm, obtains N number of pixel pair,
Each described pixel is to including that a pixel in N number of pixel and the pixel are corresponding in the second image
Matched pixel point in frame, second picture frame are a later frame image of the first image frame；
According to N number of pixel to the kinematic parameter for obtaining second picture frame, the kinematic parameter of second picture frame
It is used to indicate displacement of second picture frame relative to the first image frame；
It determines Gaussian smoothing window, includes second picture frame and second picture frame in the Gaussian smoothing window
Each M picture frame in front and back, M is positive integer；
According to M figure each before and after the second picture frame described in the beginning parameter transform model of each picture frame in the Gaussian smoothing window
As frame respectively correspond tos the relative movement parameters of second picture frame, the relative movement parameters are corresponding picture frame and institute
State the product of the kinematic parameter of each picture frame between the second picture frame；
Centered on second picture frame, according to the smooth system for assigning each picture frame in the Gaussian smoothing window in advance
Number, the relative movement parameters for respectively correspond toing second picture frame to each M picture frame before and after second picture frame carry out
Gaussian smoothing obtains the weighted average of each relative movement parameters；
The weighted average is retrieved as to the kinematic parameter of smoothed out second picture frame；
De-jitter is carried out to second picture frame according to the kinematic parameter of smoothed out second picture frame.
2. the method according to claim 1, wherein it is described according to N number of pixel to obtain described second
The kinematic parameter of picture frame, comprising:
Motion model compatible with the kinematic parameter of second picture frame is obtained, the motion model is affine model；
Determine the frequency in sampling K of random sampling unification algorism；
The calculating of random sampling unification algorism is carried out according to the affine model and the frequency in sampling K, from N number of pixel pair
Middle n pixel pair of extraction, 3≤n < N；
According to the n pixel of extraction to the kinematic parameter for calculating second picture frame with the affine model.
3. according to the method described in claim 2, it is characterized in that, the n pixel according to extraction to it is described
Affine model calculates the kinematic parameter of second picture frame, comprising:
By the n pixel to the affine model is inputted, calculated by singular value decomposition method or LU factorization described imitative
Penetrate the model parameter of model；
According to the kinematic parameter of the second picture frame described in the affine model and the model parameter calculation.
4. a kind of video image processing device, which is characterized in that described device includes:
Pixel acquisition module, in the first picture frame the N number of pixel of random acquisition, the first image frame be composition to
Handling the non-last frame image in each picture frame of video, N > 3, and N is integer；
Light stream tracing module, for carrying out light stream tracking to N number of pixel that random sampling site obtains according to LK optical flow algorithm,
Obtain N number of pixel pair, each described pixel to include N number of pixel in a pixel and the pixel
The corresponding matched pixel point in the second picture frame of point, second picture frame are a later frame image of the first image frame；
Kinematic parameter obtains module, described for the kinematic parameter according to N number of pixel to acquisition second picture frame
The kinematic parameter of second picture frame is used to indicate displacement of second picture frame relative to the first image frame；
Processing module, for carrying out de-jitter to second picture frame according to the kinematic parameter of second picture frame；
The pixel acquisition module, comprising:
Acquisition window determination unit, for determining that pixel acquisition window, the pixel acquisition window are in first figure
As the boundary of the inside of frame, and the pixel acquisition window is not overlapped with the boundary of the first image frame；
Acquisition unit, for the random acquisition pixel in the pixel acquisition window；
The acquisition window determination unit, comprising:
Window obtains subelement, for obtaining the tracking window of the light stream tracking；
Window determines subelement, and for keeping the center of the tracking window constant, the boundary of the tracking window is set
It is set to k times of primary side circle, determines that the window obtained after setting is the pixel acquisition window, k is positive number；
The processing module, comprising:
Smooth window determination unit includes second figure in the Gaussian smoothing window for determining Gaussian smoothing window
As each M picture frame before and after frame and second picture frame, M is positive integer；
Smooth unit, for the second image according to the beginning parameter transform model of each picture frame in the Gaussian smoothing window
Each M picture frame respectively correspond tos the relative movement parameters of second picture frame before and after frame, and the relative movement parameters are pair
The product of the kinematic parameter for each picture frame between picture frame and second picture frame answered；It is with second picture frame
Center, according to the smoothing factor for assigning each picture frame in the Gaussian smoothing window in advance, before second picture frame
The relative movement parameters that each M picture frame respectively correspond tos second picture frame afterwards carry out Gaussian smoothing, obtain each opposite
The weighted average of kinematic parameter；The weighted average is retrieved as to the kinematic parameter of smoothed out second picture frame；
Processing unit, for carrying out debounce to second picture frame according to the kinematic parameter of smoothed out second picture frame
Dynamic processing.
5. installation method according to claim 4, which is characterized in that the kinematic parameter obtains module, comprising:
Model acquiring unit, for obtaining motion model compatible with the kinematic parameter of second picture frame, the movement
Model is affine model；
Frequency in sampling determination unit, for determining the frequency in sampling K of random sampling unification algorism；
Extraction unit, for carrying out the calculating of random sampling unification algorism according to the affine model and the frequency in sampling K, from institute
It states N number of pixel centering and extracts n pixel pair, 3≤n < N；
Computing unit, for calculating second picture frame to the affine model according to the n pixel of extraction
Kinematic parameter.
6. device according to claim 5, which is characterized in that
The computing unit, for the n pixel to the affine model is inputted, to be passed through singular value decomposition method or LU
Decomposition method calculates the model parameter of the affine model, according to the second figure described in the affine model and the model parameter calculation
As the kinematic parameter of frame.
7. a kind of computer readable storage medium, which is characterized in that be stored with program, institute in the computer readable storage medium
Program is stated for instructing any method of video image processing of for example above-mentioned claims 1 to 3 of relevant hardware realization.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410368035.0A CN105323420B (en) | 2014-07-29 | 2014-07-29 | Method of video image processing and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410368035.0A CN105323420B (en) | 2014-07-29 | 2014-07-29 | Method of video image processing and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105323420A CN105323420A (en) | 2016-02-10 |
| CN105323420B true CN105323420B (en) | 2019-02-22 |
Family
ID=55249953
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410368035.0A Active CN105323420B (en) | 2014-07-29 | 2014-07-29 | Method of video image processing and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105323420B (en) |
Families Citing this family (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106080398A (en) * | 2016-08-27 | 2016-11-09 | 时空链（北京）科技有限公司 | A kind of automotive safety monitoring system and method |
| CN108476286B (en) * | 2016-10-17 | 2020-09-08 | 华为技术有限公司 | Image output method and electronic equipment |
| CN108881668A (en) * | 2017-06-02 | 2018-11-23 | 北京旷视科技有限公司 | Video increases steady method, apparatus, system and computer-readable medium |
| CN108307076B (en) * | 2018-03-12 | 2020-05-19 | 宁波大学 | Mobile intelligent terminal video de-jitter method |
| CN109598744B (en) * | 2018-11-29 | 2020-12-08 | 广州市百果园信息技术有限公司 | Video tracking method, device, equipment and storage medium |
| CN114387311B (en) * | 2021-12-21 | 2024-06-25 | 国能新朔铁路有限责任公司 | LKJ file and locomotive video automatic time setting method, device and computer equipment |
| CN116193257B (en) * | 2023-04-21 | 2023-09-22 | 成都华域天府数字科技有限公司 | Method for eliminating image jitter of surgical video image |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101854465A (en) * | 2010-02-01 | 2010-10-06 | 杭州海康威视软件有限公司 | Image processing method and device based on optical flow algorithm |
| CN102231792A (en) * | 2011-06-29 | 2011-11-02 | 南京大学 | Electronic image stabilization method based on characteristic coupling |
| CN102348046A (en) * | 2010-07-30 | 2012-02-08 | 富士通株式会社 | Video shake removing method and video shake removing device |
| CN102572277A (en) * | 2010-12-23 | 2012-07-11 | 三星电子株式会社 | Digital image stabilization device and method |
-
2014
- 2014-07-29 CN CN201410368035.0A patent/CN105323420B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101854465A (en) * | 2010-02-01 | 2010-10-06 | 杭州海康威视软件有限公司 | Image processing method and device based on optical flow algorithm |
| CN102348046A (en) * | 2010-07-30 | 2012-02-08 | 富士通株式会社 | Video shake removing method and video shake removing device |
| CN102572277A (en) * | 2010-12-23 | 2012-07-11 | 三星电子株式会社 | Digital image stabilization device and method |
| CN102231792A (en) * | 2011-06-29 | 2011-11-02 | 南京大学 | Electronic image stabilization method based on characteristic coupling |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105323420A (en) | 2016-02-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105323420B (en) | Method of video image processing and device | |
| US20210232924A1 (en) | Method for training smpl parameter prediction model, computer device, and storage medium | |
| CN111598998B (en) | Three-dimensional virtual model reconstruction method, three-dimensional virtual model reconstruction device, computer equipment and storage medium | |
| US11514573B2 (en) | Estimating object thickness with neural networks | |
| US20210158023A1 (en) | System and Method for Generating Image Landmarks | |
| WO2021190321A1 (en) | Image processing method and device | |
| CN109858333B (en) | Image processing method, image processing device, electronic equipment and computer readable medium | |
| CN109272509A (en) | A kind of object detection method of consecutive image, device, equipment and storage medium | |
| US20120033096A1 (en) | Motion blur modeling for image formation | |
| CN108921131B (en) | Method and device for generating face detection model and three-dimensional face image | |
| KR101925879B1 (en) | Method of tracking motion using depth image and device thereof | |
| CN110956131B (en) | Single-target tracking method, device and system | |
| CN109300151A (en) | Image processing method and device, electronic equipment | |
| WO2020107847A1 (en) | Bone point-based fall detection method and fall detection device therefor | |
| CN112561846A (en) | Method and device for training image fusion model and electronic equipment | |
| CN109447006A (en) | Image processing method, device, equipment and storage medium | |
| CN113673400A (en) | Real scene three-dimensional semantic reconstruction method and device based on deep learning and storage medium | |
| CN109345504A (en) | A kind of bottom-up more people's Attitude estimation methods constrained using bounding box | |
| CN114641799A (en) | Object detection device, method and system | |
| CN107169982A (en) | A kind of quantum LoG edge detection methods | |
| Foo et al. | Distribution-aligned diffusion for human mesh recovery | |
| CN104978583A (en) | Person action recognition method and person action recognition device | |
| CN114332927A (en) | Classroom hand-raising behavior detection method, system, computer equipment and storage medium | |
| CN114066986B (en) | Three-dimensional coordinate determination method and device, electronic equipment and storage medium | |
| CN106803056B (en) | Estimation method and device of limb relationship |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |