CN108900896A - Video clipping method and device - Google Patents
Video clipping method and device Download PDFInfo
- Publication number
- CN108900896A CN108900896A CN201810533496.7A CN201810533496A CN108900896A CN 108900896 A CN108900896 A CN 108900896A CN 201810533496 A CN201810533496 A CN 201810533496A CN 108900896 A CN108900896 A CN 108900896A
- Authority
- CN
- China
- Prior art keywords
- clip object
- target clip
- target
- model
- clipped
- 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.)
- Pending
Links
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N21/00—Selective content distribution, e.g. interactive television or video on demand [VOD]
- H04N21/40—Client devices specifically adapted for the reception of or interaction with content, e.g. set-top-box [STB]; Operations thereof
- H04N21/43—Processing of content or additional data, e.g. demultiplexing additional data from a digital video stream; Elementary client operations, e.g. monitoring of home network or synchronising decoder's clock; Client middleware
- H04N21/433—Content storage operation, e.g. storage operation in response to a pause request, caching operations
- H04N21/4334—Recording operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11B—INFORMATION STORAGE BASED ON RELATIVE MOVEMENT BETWEEN RECORD CARRIER AND TRANSDUCER
- G11B27/00—Editing; Indexing; Addressing; Timing or synchronising; Monitoring; Measuring tape travel
- G11B27/02—Editing, e.g. varying the order of information signals recorded on, or reproduced from, record carriers
- G11B27/031—Electronic editing of digitised analogue information signals, e.g. audio or video signals
- G11B27/034—Electronic editing of digitised analogue information signals, e.g. audio or video signals on discs
Abstract
This application discloses a kind of video clipping method and device, this method includes：Obtain video data to be clipped；The target clip object key frame of requirement is determined for compliance with from the video data to be clipped by established target clip object model, wherein, the target clip object model includes multiple and different target clip object associative classification model, each target clip object associative classification model for identified from the video data to be clipped with the associated corresponding key feature of target clip object so that the target clip object model determines the target clip object key frame according to multiple and different key features from the video data to be clipped；The multiple single-frame images in front and back for saving the target clip object key frame, obtain target editing video data.By the above-mentioned means, the application can provide technical support to improve the accuracy of clip object, technical support is provided for the rich of clip object.
Description
Technical field
This application involves technical field of video processing, more particularly to a kind of video clipping method, video clipping device and
Device with store function.
Background technique
Current competitive sports program, it is often necessary to splendid contents are subjected to editing, this generally requires manually to see complete field
Match, carries out editing manually, can waste many manpowers and time.
In the way of machine learning, computer is allowed to learn automatically, by video content analysis, finds out excellent portion therein
Point, excellent video content is intercepted automatically, synthesis meets the excellent collection of choice specimens of competitive sports of human demand.Such as：By language
Sound data are converted into text data, are trained using the text data after image data and conversion, establish editing model；Again
Such as：Ball and sportsman are mapped to the positioning of graphic picture coordinate system, received after image switching instructs with ball or sportsman in graphic picture
Centered on positioning on coordinate system, magnified picture is intercepted；Alternatively, establishing the classification mould after multiple training by image
Type carries out the classification based on object content to image.
But present inventor has found in long-term R&D process, the mode clip object of above-mentioned editing video
Inaccuracy, clip object are relatively simple.
Summary of the invention
The application can be the standard that improves clip object mainly solving the technical problems that provide a kind of video clipping method
True property provides technical support, provides technical support for the rich of clip object.
In order to solve the above technical problems, the technical solution that the application uses is：A kind of video clipping method, institute are provided
The method of stating includes：Obtain video data to be clipped；By established target clip object model from the video counts to be clipped
The target clip object key frame of requirement is determined for compliance in, wherein the target clip object model includes multiple and different
Target clip object associative classification model, each target clip object associative classification model are used for from the video to be clipped
Identified in data with the associated corresponding key feature of target clip object so that the target clip object model is according to more
A different key feature determines the target clip object key frame from the video data to be clipped；Described in preservation
The multiple single-frame images in front and back of target clip object key frame, obtain target editing video data.
In order to solve the above technical problems, another technical solution that the application uses is：A kind of video clipping device is provided,
Described device includes：Processor, memory and telecommunication circuit, the processor are respectively coupled to the memory and the communication
Circuit, the processor, the memory and the telecommunication circuit can be realized the step in method as described above at work
Suddenly.
In order to solve the above technical problems, another technical solution that the application uses is：It provides a kind of with store function
Device, be stored thereon with program data, described program data realize the step in method as described above when being executed by processor.
The beneficial effect of the application is：It is in contrast to the prior art, the application obtains video data to be clipped；Pass through
The target clip object that established target clip object model is determined for compliance with requirement from the video data to be clipped is crucial
Frame, wherein the target clip object model includes multiple and different target clip object associative classification models, each mesh
Mark clip object associative classification model is associated right with target clip object for identifying from the video data to be clipped
The key feature answered, so that the target clip object model is according to multiple and different key features from the view to be clipped
Frequency determines the target clip object key frame in；Save the multiple single frames figures in front and back of the target clip object key frame
Picture obtains target editing video data.Since multiple target clip object associative classification models can be from video data to be clipped
In identify with the associated multiple corresponding key features of target clip object, target clip object model is according to multiple and different
When the key feature determines target clip object key frame from video data to be clipped, can there are two types of possible mode,
One is including multiple corresponding key feature in single-frame images multiple before and after target clip object key frame, this can enable
Clip object reduces, targeted strong, the accuracy of editing is improved, by target clip object model according to multiple and different
The key feature determines target clip object key frame from video data to be clipped, provides technology to improve editing accuracy
It supports；As long as another kind is that there are any one of multiple key features in video data to be clipped, it is determined that corresponding mesh
Clip object key frame is marked, this can enable the expansions of the range of editing, and clip object is abundant, pass through target clip object model root
Target clip object key frame is determined from video data to be clipped according to multiple and different key features, to improve editing pair
The rich offer technical support of elephant.
Detailed description of the invention
In order to more clearly explain the technical solutions in the embodiments of the present application, make required in being described below to embodiment
Attached drawing is briefly described, it should be apparent that, the drawings in the following description are only some examples of the present application, 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.Wherein：
Fig. 1 is the flow diagram of one embodiment of the application video clipping method；
Fig. 2 is the flow diagram of another embodiment of the application video clipping method；
Fig. 3 is the flow diagram of the another embodiment of the application video clipping method；
Fig. 4 is that football flies into net sample schematic diagram in one specific embodiment of the application video clipping method；
Fig. 5 is that non-football flies into net sample schematic diagram in one specific embodiment of the application video clipping method；
Fig. 6 is the structural schematic diagram of one embodiment of the application video clipping device；
Fig. 7 is the structural schematic diagram for one embodiment of device that the application has store function.
Specific embodiment
Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete
Site preparation description, it is clear that described embodiments are only a part of embodiments of the present application, rather than whole embodiments.Based on this
Embodiment in application, those of ordinary skill in the art are obtained every other under the premise of not making creative labor
Embodiment shall fall in the protection scope of this application.
Refering to fig. 1, Fig. 1 is the flow diagram of one embodiment of the application video clipping method, and this method includes：
Step S101：Obtain video data to be clipped.
Step S102：Requirement is determined for compliance with from video data to be clipped by established target clip object model
Target clip object key frame, wherein target clip object model includes multiple and different target clip object associative classification mould
Type, each target clip object associative classification model are associated with for identifying from video data to be clipped with target clip object
Corresponding key feature so that target clip object model is according to multiple and different key features from video data to be clipped
Determine target clip object key frame.
Step S103：The multiple single-frame images in front and back for saving target clip object key frame, obtain target editing video counts
According to.
Video data to be clipped refers to that user wishes the video data of editing, such as：Football video, basketball video, game
Video, the match video played soccer together using one section of junior partner that mobile phone is shot, football fan or reporter are after having watched ball match
The video, etc. obtained using mobile phone video recording.
In the present embodiment, target clip object refers to the object that user wishes that editing comes out, for example, a certain specific right
As perhaps all objects associated with a certain special object perhaps the object at certain soul-stirring moment or certain have
The object, etc. of special meaning scene, these are specific require：Satisfactory target clip object.With football race
For video：Satisfactory target clip object can be this special object of scoring；Alternatively, being also possible to related to goal
A series of objects of connection are scored such as pass, the shooting before scoring, and football fan's expression excitement sound is loud and sonorous after goal, announcer explains
It has scored, sportsman and coach make celebration movement, score immediate updating, etc.；Alternatively, not scoring, also there is football fan's expression
The loud and sonorous breathtaking scene of excited sound, sportsman and coach make oiling movement, etc..Satisfactory target clip object
Key frame, as user wish key frame that editing comes out, satisfactory.
With the associated key feature of target clip object, refers to the object that foundation user wishes that editing comes out and determine
It is having on image, can judge to wish the associated feature of object that editing comes out with user with maximum probability.For example, when scoring
Carve existing key feature：There is football in picture, and football flies into net；It usually will appear these key features after goal：Football fan
Expression excitement sound is loud and sonorous, announcer's explanation has been scored, sportsman and coach make celebration movement, score immediate updating, etc..
Target clip object model is to have built up in advance, which includes multiple and different mesh
Clip object associative classification model is marked, each target clip object associative classification model can be identified from video data to be clipped
Out with the associated corresponding key feature of target clip object, multiple and different target clip object associative classification models can know
Key feature that Chu be not multiple and different, target clip object model is according to multiple and different key features from video data to be clipped
Middle determining target clip object key frame.
Wherein, target clip object model determines target according to multiple and different key features from video data to be clipped
Clip object key frame can specifically have both sides to apply, be described as follows：
On the one hand, in the prior art editing video data when, convert text data for voice data, utilize image data
It is trained with the text data after conversion, establishes editing model；By converting text for voice data among the editing model
There is distortion in the roundabout process of notebook data, the input data of editing model, there is also not for the result that final editing comes out
Accurate problem；Or only classified by the disaggregated model after image training, do not utilize other features in addition to image
Model is constructed, final editing resultant error is bigger.
A concrete application is in the application, simultaneously includes that this is more in the front and back video data of target clip object key frame
In other words a different key feature includes multiple correspondence in multiple single-frame images before and after target clip object key frame
Key feature, that is, pass through this multiple and different key feature occurred simultaneously, to analyze and determine that target clip object is crucial
Frame, it is targeted strong this can enable clip object diminution, the accuracy of editing is improved, therefore, in this way, is
It improves editing accuracy and technical support is provided.
On the other hand, the prior art is most of only by image training clip object or indirect utilization voice data, or
Person only positions sportsman and ball, and such clip object is relatively simple, and content is not enriched, cannot be simultaneously in the multiple editings of editing
Hold.
Another concrete application is in the application, as long as there are any in multiple key features in video data to be clipped
One, it is determined that corresponding target clip object key frame, this can enable the expansions of the range of editing, and clip object is abundant, lead to
It crosses target clip object model and target editing pair is determined from video data to be clipped according to multiple and different key features
As key frame, technical support is provided to improve the rich of clip object.
The multiple single-frame images in front and back for finally saving target clip object key frame, obtain target editing video data, can
So that the video data of editing is more coherent, make the content more horn of plenty of editing.
Video data to be clipped is obtained in the application embodiment；By established target clip object model from described
The target clip object key frame of requirement is determined for compliance in video data to be clipped, wherein the target clip object model packet
Multiple and different target clip object associative classification models is included, each target clip object associative classification model is used for from institute
State identified in video data to be clipped with the associated corresponding key feature of target clip object so that the target editing pair
As model determines that the target clip object is closed according to multiple and different key features from the video data to be clipped
Key frame；The multiple single-frame images in front and back for saving the target clip object key frame, obtain target editing video data.Due to more
A target clip object associative classification model can identify associated more with target clip object from video data to be clipped
A corresponding key feature, target clip object model is according to multiple and different key features from video data to be clipped
, can be there are two types of possible mode when determining target clip object key frame, one is more before and after target clip object key frame
It include multiple corresponding key feature in a single-frame images, it is targeted strong this can enable clip object diminution, it improves
The accuracy of editing, by target clip object model according to multiple and different key features from video data to be clipped
It determines target clip object key frame, provides technical support to improve editing accuracy；As long as another kind is video counts to be clipped
There are any one of multiple key features in, it is determined that corresponding target clip object key frame, this can enable cut
Volume range expand, clip object is abundant, by target clip object model according to multiple and different key features to
Target clip object key frame is determined in editing video data, provides technical support to improve the rich of clip object.
In one embodiment, key feature include occur in image target clip object, image over-the-counter sound in go out
Text appearance and in the associated instant text of target clip object, image first in existing target clip object Key Words, image
Personage occurs occurring cutting with target simultaneously with the second personage in the associated facial expression of target clip object and sound, image simultaneously
Collect at least two in the associated movement of object.
Above-mentioned key feature is the key feature often occurred in video data, generally also related to highlight, and
The more interested feature of user, can sufficiently meet user demand；On the one hand it can help improve the accuracy of editing, separately
On the one hand it also can contribute to abundant clip content.
Further, in a specific embodiment, video data to be clipped is football video to be clipped, target editing
Object is that football enters net, and key feature includes：Occur in image football enter net, occur scoring in announcer it is crucial
There is instant score in text in language, image, football fan's expression is excited in image and sound is loud and sonorous, sportsman and coach occur in image
Celebration movement.
If target clip object model has built up, can directly it use, if target clip object model is not
It establishes, then it can be first good by target clip object model foundation.In one embodiment, the foundation of target clip object model
Journey may refer to Fig. 2, and this method further includes：
Step S201：Multiple and different target clip object associative classification models is established by machine learning.
Step S202：According to multiple and different target clip object associative classification model foundation target clip object models.
Machine learning (Machine Learning, ML) is a multi-field cross discipline, be related to probability theory, statistics,
The multiple subjects such as Approximation Theory, convextiry analysis, algorithm complexity theory；Specialize in the study that the mankind were simulated or realized to computer how
Behavior reorganizes the existing structure of knowledge and is allowed to constantly improve the performance of itself to obtain new knowledge or skills；It is people
The core of work intelligence, is the fundamental way for making computer have intelligence, and application spreads the every field of artificial intelligence.
In present embodiment, multiple and different target clip object associative classification models is established by machine learning, then
Further according to the model that multiple and different target clip object associative classification model foundations is final, i.e. target clip object model, lead to
This mode is crossed, the time of building target clip object model can be substantially reduced, and can intelligently improve the performance of itself, is made
Final editing result is more met the needs of users.
Further, in step s 201, multiple and different target clip object associative classification moulds is established by machine learning
Type can specifically include：Establish multiple and different target clip object associative classification models respectively using AdaBoost classifier.
Adaboost is a kind of iterative algorithm, and core concept is the classifier different for the training of the same training set
(Weak Classifier) then gets up these weak classifier sets, constitutes a stronger final classification device (strong classifier).It is calculated
Method itself is realized by changing data distribution, and whether it is correct according to the classification of each sample among each training set,
And the accuracy rate of the general classification of last time, to determine the weight of each sample.The new data set for modifying weight is given down
Layer classifier is trained, the Multiple Classifier Fusion for finally obtaining each training, as last Decision Classfication device.It uses
Adaboost classifier can exclude some unnecessary training data features, and be placed on above crucial training data.
General machine-learning process include sample data acquisition, sample production, sample characteristics extraction, classifier design,
The processes such as data training, parameter adjustment, test result feedback, actual test.The application in one embodiment, picture training sample
Original training, can simplify machine-learning process.Fig. 3 is referred to, is established respectively using AdaBoost classifier multiple and different
Target clip object associative classification model, can specifically include：
Step S301：Obtain and mark predetermined quantity, the picture training sample including there is key feature.
Step S302：Sample characteristics extraction is carried out to picture training sample, establishes the feature vector square of picture training sample
Battle array.
Step S303：By the eigenvectors matrix input AdaBoost classifier of picture training sample to be trained, obtain
To not after tested, meet the initial target clip object associative classification model of preset requirement.
Step S304：Using the video measurement test sample including there is key feature and adjust initial target editing
Object associative classification model obtains the target clip object associative classification model for meeting preset requirement.
The above process can identify that football flies into the key feature of net and is with target clip object associative classification model
Example, to be described in detail：
(1) data acquisition and sample manufacturing process：It acquires a large amount of football and flies into net picture as shown in figure 4, these are drawn
Occur football and net in face, and football has flown into net, while a large amount of non-footballs also being taken to fly into net picture, such as Fig. 5
It is shown, do not occur net or football in these samples, the feature of net is also entered without football.Picture training will be collected
Sample carries out dimension normalization and label.
(2) sample characteristics extraction process：Football, which can be summed up, by observation chart 4 and the feature of image of Fig. 5 flies into net
Key features：Picture has football, also there is a net, and the distance of football and net is closer, and being reflected in two dimensional image is exactly football
Pixel and net pixel apart from closer, can be by color and straight in addition to this there are also the doorframe and door line of white
Line detects frame out and white door line and doorframe occurs.Football available in this way flies into the sample characteristics of net, can build later
Vertical eigenvectors matrix (is infused as shown in table 1：Numerical value in eigenmatrix vector is in order to illustrate method, with actual measurement data
Have deviation), it can intuitively find out that different feature combinations indicates that the football of the picture flies into net according to eigenvectors matrix
Probability.
1 eigenvectors matrix of table
(3) AdaBoost classifier basic principle：AdaBoost classifier Theory comparison is mature, and in Face datection and knows
Effective practice is obtained in other isotype identification classification.This method allows to continually add new " Weak Classifier ", until reaching
Some scheduled sufficiently small error rate.In AdaBoost method, each picture training sample is endowed a weight,
Show that it is selected into the probability of training set by some classifier.If some sample is by Accurate classification, under construction one
In a training set, its selected probability is just lowered；On the contrary, if some sample is not classified correctly, its power
Weight is just improved.In this way, AdaBoost method " can focus on " more difficult (the richer letter of those classification
Breath) sample on.These weak detectors are only better than random guess, and the conjecture for two class problems only than 50% is good
A bit.But the very weak Multiple Classifier Fusion of these detectabilities is got up by certain algorithm, a classification capacity will be obtained
Very strong strong classifier.It should be noted that being not limited to AdaBoost classifier here, support vector machines also can choose
The others such as (Support Vector Machine, SVM) classifier, it will not be described here.
The algorithm flow of AdaBoost：
Input：Training dataset T={ (x1,y1),(x2,y2)......(xN,yN), whereiny∈Y
=-1,1, iteration total degree M；
1, the weight distribution of training sample is initialized：
2, for the number of iterations m=1,2 ..., M：
(a) D is distributed using with weightmTraining dataset learnt, obtain Weak Classifier Gm(x)；
(b) G is calculatedm(x) D is distributed in training data centralized valuemOn error in classification rate em：
(c) G is calculatedm(x) weight α shared in final classification devicem：
(d) weight for updating training dataset is distributed (z heremNormalization factor, in order to make sample probability distribution and
For 1)：
3, final classification device is obtained：
(4) training and adjustment process：Picture training sample is divided into training sample and test sample, training sample is mainly
For carrying out classifier study, test sample is mainly to detect whether classification learning parameter meets the requirements.It first will training sample
In this feeding classifier, according to the process of classifier, the iterative characteristic extraction of data is carried out, characteristic parameter compares, the spy of iteration
Sign parametric classification threshold calculations, sample such as reclassify at the processes.Later, by the calculated result parameter of these processes in test specimens
Characteristic vector pickup, characteristic parameter sample are carried out in sheet and the processes such as reclassifies, and finally obtain the accuracy and mistake of sample judgement
Accidentally rate, if accuracy and error rate meet preset requirement, for example require the correct probability of classification 95% or more, then classifying
Device study is completed；, whereas if test result accuracy is lower than 95%, then to readjust the parameter setting of classifier or add
New characteristic attribute etc. is added in the quantity of large sample.
(5) actual test process：Above-mentioned classification only completes learning test process on limited sample set, one at
The classifier of function also needs to be tested in real data, and the calculated result parameter of these processes is enterprising in real data
Row characteristic vector pickup, characteristic parameter sample such as reclassify at the processes, finally obtain the accuracy and error rate of sample judgement, such as
Fruit accuracy and error rate meet preset requirement, for example require the correct probability of classification 95% or more, then classifier
It practises and completing；, whereas if test result accuracy is lower than 95%, then to readjust the parameter setting of classifier or increase sample
New characteristic attribute etc. is added in this quantity.
In one embodiment, in step S202, according to multiple and different target clip object associative classification model foundations
Target clip object model, can specifically include：
Based on decision-tree model, joins together multiple and different target clip object associative classification models to establish target and cut
Collect object model.
Decision tree (Decision Tree) is a kind of simple but widely used classifier；It is constructed by training data
Decision tree can efficiently classify to unknown data.Decision number has two big advantages：(1) decision-tree model can the property read
It is good, have descriptive, facilitates manual analysis；(2) high-efficient, decision tree only needs once to construct, Reusability, each time in advance
The max calculation number of survey is no more than the depth of decision tree.
It is the structural schematic diagram of one embodiment of the application video clipping device referring to Fig. 6, Fig. 6, which includes：Processing
Device 1, memory 2 and telecommunication circuit 3, processor 1 are respectively coupled to memory 2 and telecommunication circuit 3, processor 1, memory 2 with
And telecommunication circuit 3 can be realized the step in as above any one the method at work.The detailed description of related content please be joined
See above method part, it is no longer superfluous herein to chat.
It is the structural schematic diagram for one embodiment of device that the application has store function, the device 10 referring to Fig. 7, Fig. 7
On be stored with program data 100, the step any one of as above in method is realized when program data 100 is executed by processor.It is related
The detailed description of content refers to above method part, no longer superfluous herein to chat.
The technical solution of the application substantially the part that contributes to existing technology or the technical solution in other words
Completely or partially can be embodied in the form of software products, this has the device of store function, including several program datas,
With so that a computer equipment (can be personal computer, server or the network equipment etc.) or processor
(processor) all or part of the steps of each embodiment the method for the application is executed.And it is above-mentioned with storage function
Can device include：USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory
(RAM, Random Access Memory), etc..
The foregoing is merely presently filed embodiments, are not intended to limit the scope of the patents of the application, all to utilize this
Equivalent structure or equivalent flow shift made by application specification and accompanying drawing content, it is relevant to be applied directly or indirectly in other
Technical field similarly includes in the scope of patent protection of the application.
Claims (10)
1. a kind of video clipping method, which is characterized in that the method includes：
Obtain video data to be clipped；
The target editing of requirement is determined for compliance with from the video data to be clipped by established target clip object model
Object key frame, wherein the target clip object model includes multiple and different target clip object associative classification model, often
A target clip object associative classification model from the video data to be clipped for identifying and target clip object
Associated corresponding key feature, so that the target clip object model is according to multiple and different key features from described
The target clip object key frame is determined in video data to be clipped；
The multiple single-frame images in front and back for saving the target clip object key frame, obtain target editing video data.
2. the method according to claim 1, wherein the key feature includes occurring target editing pair in image
As, image over-the-counter sound in there are target clip object Key Words, being associated with target clip object occurs in the text in image
Instant text, the first personage occurs and in the associated facial expression of target clip object and sound, image the simultaneously in image
Two personages occur and at least two in the associated movement of target clip object simultaneously.
3. according to the method described in claim 2, it is characterized in that, the video data to be clipped is that football to be clipped regards
Frequently, the target clip object is that football enters net, and the key feature includes：Occur football in image and enters net, solution
Occur goal Key Words in the person of saying, instant score occurs in the text in image, football fan's expression is excited in image and sound is loud and sonorous, figure
There is celebration movement in sportsman and coach as in.
4. the method according to claim 1, wherein the method also includes：
Multiple and different target clip object associative classification models is established by machine learning；
According to multiple and different target clip object associative classification model foundation target clip object models.
5. according to the method described in claim 4, it is characterized in that, described establish multiple and different targets by machine learning and cut
Object associative classification model is collected, including：
Establish multiple and different target clip object associative classification models respectively using AdaBoost classifier.
6. according to the method described in claim 5, it is characterized in that, it is described using AdaBoost classifier establish respectively it is multiple not
Same target clip object associative classification model, including：
Obtain and mark predetermined quantity, the picture training sample including there is key feature；
Sample characteristics extraction is carried out to the picture training sample, establishes the eigenvectors matrix of the picture training sample；
The eigenvectors matrix of the picture training sample is inputted into the AdaBoost classifier to be trained, obtain without
Test, meet the initial target clip object associative classification model of preset requirement；
Using the video measurement test sample including there is key feature and adjust the initial target clip object association
Disaggregated model obtains the target clip object associative classification model for meeting preset requirement.
7. according to the method described in claim 6, it is characterized in that, the preset requirement includes default accuracy and default mistake
Rate.
8. according to the method described in claim 4, it is characterized in that, described close according to multiple and different target clip objects
Connection disaggregated model establishes target clip object model, including：
Based on decision-tree model, join together multiple and different target clip object associative classification models to establish the mesh
Mark clip object model.
9. a kind of video clipping device, which is characterized in that described device includes：Processor, memory and telecommunication circuit, it is described
Processor is respectively coupled to the memory and the telecommunication circuit, the processor, the memory and the telecommunication circuit
It can be realized the step in any one of claim 1-8 the method at work.
10. a kind of device with store function, is stored thereon with program data, which is characterized in that described program data are located
Manage the step realized in any one of claim 1-8 the method when device executes.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810533496.7A CN108900896A (en) | 2018-05-29 | 2018-05-29 | Video clipping method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810533496.7A CN108900896A (en) | 2018-05-29 | 2018-05-29 | Video clipping method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108900896A true CN108900896A (en) | 2018-11-27 |
Family
ID=64343528
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810533496.7A Pending CN108900896A (en) | 2018-05-29 | 2018-05-29 | Video clipping method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108900896A (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111836100A (en) * | 2019-04-16 | 2020-10-27 | 阿里巴巴集团控股有限公司 | Method, apparatus, device and storage medium for creating clip track data |
| CN111914102A (en) * | 2020-08-27 | 2020-11-10 | 上海掌门科技有限公司 | Method for editing multimedia data, electronic device and computer storage medium |
| CN112218005A (en) * | 2020-09-23 | 2021-01-12 | 深圳锐取信息技术股份有限公司 | Video editing method based on artificial intelligence |
| CN112990142A (en) * | 2021-04-30 | 2021-06-18 | 平安科技（深圳）有限公司 | Video guide generation method, device and equipment based on OCR (optical character recognition), and storage medium |
| CN113490049A (en) * | 2021-08-10 | 2021-10-08 | 深圳市前海动竞体育科技有限公司 | Sports event video editing method and system based on artificial intelligence |
| CN114302224A (en) * | 2021-12-23 | 2022-04-08 | 新华智云科技有限公司 | Intelligent video editing method, device, equipment and storage medium |
| CN114422851A (en) * | 2022-01-24 | 2022-04-29 | 腾讯科技（深圳）有限公司 | Video clipping method, video clipping device, electronic equipment and readable medium |
| WO2022184117A1 (en) * | 2021-03-04 | 2022-09-09 | 腾讯科技（深圳）有限公司 | Deep learning-based video clipping method, related device, and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002080027A1 (en) * | 2001-03-29 | 2002-10-10 | British Telecommunications Public Limited Company | Image processing |
| WO2006132596A1 (en) * | 2005-06-07 | 2006-12-14 | Matsushita Electric Industrial Co., Ltd. | Method and apparatus for audio clip classification |
| CN102427507A (en) * | 2011-09-30 | 2012-04-25 | 北京航空航天大学 | Football video highlight automatic synthesis method based on event model |
| CN105516651A (en) * | 2014-10-14 | 2016-04-20 | 韩华泰科株式会社 | Method and apparatus for providing combined-summary in imaging apparatus |
| CN105912560A (en) * | 2015-02-24 | 2016-08-31 | 泽普实验室公司 | Detect sports video highlights based on voice recognition |
| CN106028134A (en) * | 2015-03-31 | 2016-10-12 | 泽普实验室公司 | Detect sports video highlights for mobile computing devices |
| CN106534967A (en) * | 2016-10-25 | 2017-03-22 | 司马大大（北京）智能系统有限公司 | Video editing method and device |
| CN107194419A (en) * | 2017-05-10 | 2017-09-22 | 百度在线网络技术（北京）有限公司 | Video classification methods and device, computer equipment and computer-readable recording medium |
| CN107566907A (en) * | 2017-09-20 | 2018-01-09 | 广东欧珀移动通信有限公司 | video clipping method, device, storage medium and terminal |
-
2018
- 2018-05-29 CN CN201810533496.7A patent/CN108900896A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2002080027A1 (en) * | 2001-03-29 | 2002-10-10 | British Telecommunications Public Limited Company | Image processing |
| WO2006132596A1 (en) * | 2005-06-07 | 2006-12-14 | Matsushita Electric Industrial Co., Ltd. | Method and apparatus for audio clip classification |
| CN102427507A (en) * | 2011-09-30 | 2012-04-25 | 北京航空航天大学 | Football video highlight automatic synthesis method based on event model |
| CN105516651A (en) * | 2014-10-14 | 2016-04-20 | 韩华泰科株式会社 | Method and apparatus for providing combined-summary in imaging apparatus |
| CN105912560A (en) * | 2015-02-24 | 2016-08-31 | 泽普实验室公司 | Detect sports video highlights based on voice recognition |
| CN106028134A (en) * | 2015-03-31 | 2016-10-12 | 泽普实验室公司 | Detect sports video highlights for mobile computing devices |
| CN106534967A (en) * | 2016-10-25 | 2017-03-22 | 司马大大（北京）智能系统有限公司 | Video editing method and device |
| CN107194419A (en) * | 2017-05-10 | 2017-09-22 | 百度在线网络技术（北京）有限公司 | Video classification methods and device, computer equipment and computer-readable recording medium |
| CN107566907A (en) * | 2017-09-20 | 2018-01-09 | 广东欧珀移动通信有限公司 | video clipping method, device, storage medium and terminal |
Non-Patent Citations (1)
| Title |
|---|
| 许荣: "融合音视频特征的足球视频检索研究", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111836100A (en) * | 2019-04-16 | 2020-10-27 | 阿里巴巴集团控股有限公司 | Method, apparatus, device and storage medium for creating clip track data |
| CN111914102A (en) * | 2020-08-27 | 2020-11-10 | 上海掌门科技有限公司 | Method for editing multimedia data, electronic device and computer storage medium |
| CN112218005A (en) * | 2020-09-23 | 2021-01-12 | 深圳锐取信息技术股份有限公司 | Video editing method based on artificial intelligence |
| WO2022184117A1 (en) * | 2021-03-04 | 2022-09-09 | 腾讯科技（深圳）有限公司 | Deep learning-based video clipping method, related device, and storage medium |
| CN112990142A (en) * | 2021-04-30 | 2021-06-18 | 平安科技（深圳）有限公司 | Video guide generation method, device and equipment based on OCR (optical character recognition), and storage medium |
| CN113490049A (en) * | 2021-08-10 | 2021-10-08 | 深圳市前海动竞体育科技有限公司 | Sports event video editing method and system based on artificial intelligence |
| CN113490049B (en) * | 2021-08-10 | 2023-04-21 | 深圳市前海动竞体育科技有限公司 | Sports event video editing method and system based on artificial intelligence |
| CN114302224A (en) * | 2021-12-23 | 2022-04-08 | 新华智云科技有限公司 | Intelligent video editing method, device, equipment and storage medium |
| CN114302224B (en) * | 2021-12-23 | 2023-04-07 | 新华智云科技有限公司 | Intelligent video editing method, device, equipment and storage medium |
| CN114422851A (en) * | 2022-01-24 | 2022-04-29 | 腾讯科技（深圳）有限公司 | Video clipping method, video clipping device, electronic equipment and readable medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108900896A (en) | Video clipping method and device | |
| CN109359215B (en) | Video intelligent pushing method and system | |
| Zhu et al. | Visual7w: Grounded question answering in images | |
| CN110008842A (en) | A kind of pedestrian's recognition methods again for more losing Fusion Model based on depth | |
| WO2019228268A1 (en) | Method for displaying live broadcast room, apparatus, device, and storage medium | |
| CN110162593A (en) | A kind of processing of search result, similarity model training method and device | |
| CN102184221B (en) | Real-time video abstract generation method based on user preferences | |
| US20230049135A1 (en) | Deep learning-based video editing method, related device, and storage medium | |
| CN105183849B (en) | A kind of match Video Events detection of Snooker and semanteme marking method | |
| CN108491443A (en) | By computer-implemented and method and computer system of user session | |
| CN106778852A (en) | A kind of picture material recognition methods for correcting erroneous judgement | |
| CN109344884A (en) | The method and device of media information classification method, training picture classification model | |
| CN107463698A (en) | Method and apparatus based on artificial intelligence pushed information | |
| CN112132197A (en) | Model training method, image processing method, device, computer equipment and storage medium | |
| CN111400536B (en) | Low-cost tomato leaf disease identification method based on lightweight deep neural network | |
| JP6366626B2 (en) | Generating device, generating method, and generating program | |
| CN111160134A (en) | Human-subject video scene analysis method and device | |
| CN110652726B (en) | Game auxiliary system based on image recognition and audio recognition | |
| CN109961097A (en) | Image classification dispatching method based on edge calculations under a kind of embedded scene | |
| WO2018033066A1 (en) | Robot control method and companion robot | |
| CN108038601A (en) | Teaching methods of physical education, device and computer-readable recording medium | |
| CN108345612A (en) | A kind of question processing method and device, a kind of device for issue handling | |
| CN110610500A (en) | News video self-adaptive strip splitting method based on dynamic semantic features | |
| CN111282281B (en) | Image processing method and device, electronic equipment and computer readable storage medium | |
| CN109242309A (en) | The user that attends a meeting portrait generation method, device, intelligent meeting equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20181127 |