CN110163079A - Video detecting method and device, computer-readable medium and electronic equipment - Google Patents
Video detecting method and device, computer-readable medium and electronic equipment Download PDFInfo
- Publication number
- CN110163079A CN110163079A CN201910226798.4A CN201910226798A CN110163079A CN 110163079 A CN110163079 A CN 110163079A CN 201910226798 A CN201910226798 A CN 201910226798A CN 110163079 A CN110163079 A CN 110163079A
- Authority
- CN
- China
- Prior art keywords
- video
- cryptographic hash
- global characteristics
- detected
- image
- 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
- 238000000034 method Methods 0.000 title claims abstract description 81
- 238000003860 storage Methods 0.000 claims abstract description 36
- 238000001514 detection method Methods 0.000 claims abstract description 35
- 238000012545 processing Methods 0.000 claims abstract description 18
- 238000012360 testing method Methods 0.000 claims abstract description 12
- 238000013527 convolutional neural network Methods 0.000 claims description 64
- 238000012549 training Methods 0.000 claims description 38
- 238000006243 chemical reaction Methods 0.000 claims description 21
- 238000007667 floating Methods 0.000 claims description 16
- 238000002360 preparation method Methods 0.000 claims description 10
- 238000013139 quantization Methods 0.000 claims description 10
- 238000004590 computer program Methods 0.000 claims description 6
- 239000000284 extract Substances 0.000 abstract description 8
- 230000006870 function Effects 0.000 description 69
- 230000008569 process Effects 0.000 description 26
- 238000010586 diagram Methods 0.000 description 22
- 238000000605 extraction Methods 0.000 description 13
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000006854 communication Effects 0.000 description 5
- 238000004891 communication Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 230000009466 transformation Effects 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000013461 design Methods 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000006399 behavior Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 230000005291 magnetic effect Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 208000006650 Overbite Diseases 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000006978 adaptation Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 230000003014 reinforcing effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 150000003431 steroids Chemical class 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computational Linguistics (AREA)
- Multimedia (AREA)
- Software Systems (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of video detecting method and devices, computer-readable medium and electronic equipment, are related to technical field of image processing.The video detecting method includes: preparatory acquisition video frame sample, determine the first global characteristics and the second global characteristics of video frame sample, second global characteristics are converted into the first cryptographic Hash, is established and is indexed using the first cryptographic Hash, and the first global characteristics are stored under index;It determines the first global characteristics and the second global characteristics of key frame of video to be detected, and the second global characteristics of key frame of video to be detected is converted into the second cryptographic Hash;When the similarity of the second cryptographic Hash and the first cryptographic Hash is greater than the first similarity threshold, extract corresponding first global characteristics of the second cryptographic Hash, corresponding first global characteristics of second cryptographic Hash are compared with the global characteristics of the corresponding storage of the first cryptographic Hash, and determine the testing result of video to be detected according to comparison result.The disclosure can quickly and accurately realize video detection.
Description
Technical field
This disclosure relates to technical field of image processing, in particular to a kind of video detecting method, device, computer
Readable medium and electronic equipment.
Background technique
With the continuous development of machine learning techniques, machine learning is widely used in every field.For example, will
Machine learning techniques are applied to field of video detection, for video to be detected, therefrom extract image, computer system can be certainly
The image for being about to extract is compared with other images, and then can determine image similar with the image, and according to image
Comparison result, complete the detection of video.
Currently, the disaggregated model based on convolutional neural networks can effectively detect image, however, for each
The process of detection, process is complicated and takes a long time, can not be quickly and accurately to figure especially under the scene of detection video image
As being detected, lead to not preferably obtain video detection result.
It should be noted that information is only used for reinforcing the reason to the background of the disclosure disclosed in above-mentioned background technology part
Solution, therefore may include the information not constituted to the prior art known to persons of ordinary skill in the art.
Summary of the invention
The disclosure is designed to provide a kind of video detecting method, device, computer-readable medium and electronic equipment, into
And it can not be quickly and accurately to image caused by overcoming the limitation and defect due to the relevant technologies at least to a certain extent
The problem of being detected.
According to one aspect of the disclosure, a kind of video detecting method is provided, comprising: video frame sample is obtained in advance,
Second global characteristics are converted to the first cryptographic Hash by the first global characteristics and the second global characteristics for determining video frame sample, benefit
It is established and is indexed with the first cryptographic Hash, and the first global characteristics are stored under index;Extract the video to be detected of video to be detected
Key frame, determines the first global characteristics and the second global characteristics of key frame of video to be detected, and by key frame of video to be detected
The second global characteristics be converted to the second cryptographic Hash;When the similarity of the second cryptographic Hash and the first cryptographic Hash is greater than the first similarity
When threshold value, corresponding first global characteristics of the second cryptographic Hash are extracted, by corresponding first global characteristics of the second cryptographic Hash and first
The global characteristics of the corresponding storage of cryptographic Hash are compared, and the testing result of video to be detected is determined according to comparison result.
According to one aspect of the disclosure, a kind of video detecting device, including detection preparation module, feature extraction are provided
Module and feature comparison module.
Specifically, detection preparation module determines that the first of video frame sample is global special for obtaining video frame sample in advance
It seeks peace the second global characteristics, the second global characteristics is converted into the first cryptographic Hash, established and indexed using the first cryptographic Hash, and by the
One global characteristics are stored under index;Characteristic extracting module is used to extract the key frame of video to be detected of video to be detected, determines
The first global characteristics and the second global characteristics of key frame of video to be detected, and it is global special by the second of key frame of video to be detected
Sign is converted to the second cryptographic Hash;Feature comparison module is used to be greater than the first phase when the similarity of the second cryptographic Hash and the first cryptographic Hash
When like degree threshold value, extract corresponding first global characteristics of the second cryptographic Hash, by corresponding first global characteristics of the second cryptographic Hash with
The global characteristics of the corresponding storage of first cryptographic Hash are compared, and the testing result of video to be detected is determined according to comparison result.
Optionally, the first global characteristics and the second global characteristics of video frame are determined using convolutional neural networks, wherein
Global characteristics are full articulamentum feature.
Specifically, detection preparation module determines that the first of the video frame sample is complete for obtaining video frame sample in advance
Described second full articulamentum Feature Conversion is the first cryptographic Hash, using described by articulamentum feature and the second full articulamentum feature
First cryptographic Hash establishes index, and the floating point features of the described first full articulamentum is stored under the index;Feature extraction mould
Block is used to extract the key frame of video to be detected of video to be detected, determines the first full articulamentum of the key frame of video to be detected
Floating point features and the second full articulamentum feature, and be the by the second of the key frame of video to be detected the full articulamentum Feature Conversion
Two cryptographic Hash;Feature comparison module is used to be greater than first to the similarity of first cryptographic Hash when second cryptographic Hash similar
When spending threshold value, the storage corresponding with first cryptographic Hash of the corresponding first full articulamentum floating point features of second cryptographic Hash is extracted
Full articulamentum floating point features be compared, and the testing result of the video to be detected is determined according to comparison result.
Optionally, video frame sample include positive sample to and negative sample;Wherein, detection preparation module further includes sample acquisition
Unit.
Specifically, sample acquisition unit is configured as executing: obtaining positive sample pair in advance, positive sample is to including the first image
With the second image;Wherein, the first image is identical as the classification of the second image;It is different from the classification of positive sample pair that classification is obtained in advance
Third image, as negative sample;Wherein, using the positive sample to and negative sample the convolutional neural networks are instructed
Practice, to adjust the parameter in the convolutional neural networks.
Optionally, sample acquisition unit is additionally configured to execute: an image is obtained from video sample, as the first figure
Picture;Using in video sample away from the image in the first image preset time period as the second image.
Optionally, sample acquisition unit is additionally configured to execute: an image is obtained from video sample, as the first figure
Picture;Form conversion is carried out to the first image, to determine the second image.
Optionally, sample acquisition unit is additionally configured to execute: obtaining an image from video sample, waits as first
Select image;Using in video sample away from the image in the first candidate image preset time period as the second candidate image;Respectively to
One candidate image and the second candidate image carry out form conversion, to determine candidate image collection;It is concentrated from candidate image and obtains two
Image, respectively as the first image and the second image.
Optionally, characteristic extracting module includes network training unit and feature extraction unit.
Specifically, network training unit is used to video frame sample inputting convolutional neural networks, convolutional neural networks are calculated
Loss function, restrain the loss function of convolutional neural networks, with determine training after convolutional neural networks;Feature extraction
Unit is used for the convolutional neural networks after the key frame of video to be detected input training by video to be detected, will connect entirely via first
The first global characteristics of feature that layer is determined as key frame of video to be detected are connect, and will be determined via the second full articulamentum
Second global characteristics of the feature as key frame of video to be detected.
Optionally, the loss function of convolutional neural networks includes range loss function, quantization loss function and comentropy damage
Lose function;
Range loss function are as follows: Lt=max { (xa-xp)+m-(xa-xn),0};
Quantify loss function are as follows: Lq=| | | x(i)|-1||1;
Comentropy loss function are as follows:
Wherein, LtIndicate range loss function, xa、xp、xnRespectively indicate the spy of two samples and negative sample of positive sample centering
Sign, m indicate the confidence level of feature spacing;LqIndicate quantization loss function, x(i)Indicate any feature of video frame sample;LeIt indicates
Comentropy loss function, udIndicate the mean value of the feature of video frame sample.
Optionally, characteristic extracting module further includes key-frame extraction unit.
Specifically, key-frame extraction unit is used to extract video frame from video to be detected every prefixed time interval, make
For key frame of video to be detected.
Optionally, video detecting device further includes copyright detection module.
Specifically, copyright detection module is configured as executing: if corresponding first global characteristics of second cryptographic Hash
The similarity of global characteristics corresponding with first cryptographic Hash is greater than the second similarity threshold, it is determined that the video to be detected
Upload side mark and first cryptographic Hash correspond to video upload side mark, when the video to be detected upload side mark
When the upload side that knowledge and first cryptographic Hash correspond to video identifies different, send a warning message.
Optionally, video detecting device further includes label adding module.
Specifically, label adding module is configured as executing: if corresponding first global characteristics of second cryptographic Hash
The similarity of global characteristics corresponding with first cryptographic Hash is greater than third similarity threshold, it is determined that first cryptographic Hash
The label of corresponding video, and the label that first cryptographic Hash corresponds to video is added to the video to be detected.
Optionally, video detecting device further includes video uploading module.
Specifically, video uploading module is configured as executing: if corresponding first global characteristics of second cryptographic Hash
The similarity of global characteristics corresponding with first cryptographic Hash, then will be on the video to be detected less than the 4th similarity threshold
It passes.
According to one aspect of the disclosure, a kind of computer-readable medium is provided, computer program is stored thereon with, is counted
Any one of the above video detecting method when calculation machine program is executed by processor.
According to one aspect of the disclosure, a kind of electronic equipment is provided, comprising: one or more processors;Storage device,
For storing one or more programs, when one or more programs are executed by one or more processors, so that one or more
A processor realizes such as any one of the above video detecting method.
In the technical solution provided by some embodiments of the present disclosure, firstly, global using the first of video frame sample
Feature construction video fingerprint data library utilizes the rope of the second global characteristics building and video fingerprint data library of video frame sample
Draw;Next, the first global characteristics and the second global characteristics of key frame of video to be detected are extracted, the index based on building and benefit
It determines to be compared with the first global characteristics of key frame of video to be detected with the second global characteristics of key frame of video to be detected
Pair feature, and aspect ratio pair is carried out, to determine testing result.The illustrative embodiments of the disclosure are by providing index function
Can, can quickly detect may global characteristics relevant to video to be detected, then it is special using the overall situation of video to be detected itself
Sign is compared with possible relevant global characteristics, can accurately determine out and whether there is view similar with video to be detected
Frequently.
It should be understood that above general description and following detailed description be only it is exemplary and explanatory, not
The disclosure can be limited.
Detailed description of the invention
The drawings herein are incorporated into the specification and forms part of this specification, and shows the implementation for meeting the disclosure
Example, and together with specification for explaining the principles of this disclosure.It should be evident that the accompanying drawings in the following description is only the disclosure
Some embodiments for those of ordinary skill in the art without creative efforts, can also basis
These attached drawings obtain other attached drawings.In the accompanying drawings:
Fig. 1 is shown can be using the video detecting method of the embodiment of the present disclosure or the exemplary system of video detecting device
The schematic diagram of framework;
Fig. 2 shows the structural schematic diagrams of the computer system of the electronic equipment suitable for being used to realize the embodiment of the present disclosure;
Fig. 3 diagrammatically illustrates the flow chart of the video detecting method according to disclosure illustrative embodiments;
Fig. 4 diagrammatically illustrates the network of the convolutional neural networks according to used in the training of disclosure illustrative embodiments
Structure chart;
Fig. 5 shows the schematic diagram of the determination positive sample pair according to a kind of illustrative embodiments of the disclosure;
Fig. 6 shows the schematic diagram of the determination positive sample pair according to disclosure another kind illustrative embodiments;
Fig. 7, which is diagrammatically illustrated, uses the video detecting method of disclosure illustrative embodiments to generate label for video
Schematic diagram;
Fig. 8 is shown under bonding wire according to an exemplary embodiment of the present disclosure and scheme realizes video detection on line
Schematic diagram;
Fig. 9 diagrammatically illustrates the block diagram of video detecting device according to an exemplary embodiment of the present disclosure;
Figure 10 diagrammatically illustrates the block diagram of detection preparation module according to an exemplary embodiment of the present disclosure;
Figure 11 diagrammatically illustrates the block diagram of characteristic extracting module according to an exemplary embodiment of the present disclosure;
Figure 12 diagrammatically illustrates the block diagram of the characteristic extracting module of the another exemplary embodiment according to the disclosure;
Figure 13 diagrammatically illustrates the block diagram of the video detecting device of the another exemplary embodiment according to the disclosure;
Figure 14 diagrammatically illustrates the block diagram of the video detecting device of the another exemplary embodiment according to the disclosure;
Figure 15 diagrammatically illustrates the block diagram of the video detecting device of another illustrative embodiments according to the disclosure.
Specific embodiment
Example embodiment is described more fully with reference to the drawings.However, example embodiment can be with a variety of shapes
Formula is implemented, and is not understood as limited to example set forth herein;On the contrary, thesing embodiments are provided so that the disclosure will more
Fully and completely, and by the design of example embodiment comprehensively it is communicated to those skilled in the art.Described feature, knot
Structure or characteristic can be incorporated in any suitable manner in one or more embodiments.In the following description, it provides perhaps
More details fully understand embodiment of the present disclosure to provide.It will be appreciated, however, by one skilled in the art that can
It is omitted with technical solution of the disclosure one or more in the specific detail, or others side can be used
Method, constituent element, device, step etc..In other cases, be not shown in detail or describe known solution to avoid a presumptuous guest usurps the role of the host and
So that all aspects of this disclosure thicken.
In addition, attached drawing is only the schematic illustrations of the disclosure, it is not necessarily drawn to scale.Identical attached drawing mark in figure
Note indicates same or similar part, thus will omit repetition thereof.Some block diagrams shown in the drawings are function
Energy entity, not necessarily must be corresponding with physically or logically independent entity.These function can be realized using software form
Energy entity, or these functional entitys are realized in one or more hardware modules or integrated circuit, or at heterogeneous networks and/or place
These functional entitys are realized in reason device device and/or microcontroller device.
Flow chart shown in the drawings is merely illustrative, it is not necessary to including all steps.For example, the step of having
It can also decompose, and the step of having can merge or part merges, therefore the sequence actually executed is possible to according to the actual situation
Change.It should be understood that term " first ", " second ", " third ", " the 4th " etc. are merely to the purpose distinguished, should not make
For the limitation of disclosure corresponding content.
It is detected to image, more particularly under the scene detected to video image, some technologies use base
In the scheme of VGG (Visual Geometry Group, visual geometric group) network integration sparse coding (Sparse Coding),
However, this scheme needs intensive characteristic extraction procedure, time complexity is higher, and process is complicated, can not be preferably extensive
It is applied in complicated video detection task;Other technologies use the unsupervised Hash based on DeepBit (depth bit)
Algorithm, this kind of scheme is using Hash feature as the feature (or be fingerprint) of final image (video), however, by floating point features
It is converted into Hash procedure, some effective informations can be lost, cause operation result inaccurate, in addition, for complexity and possible
The video scene of transformation, this kind of scheme can not also be applied on a large scale.
In consideration of it, disclosure illustrative embodiments provide a kind of video detecting method and video detecting device, with solution
The certainly above problem.
The embodiment of the present disclosure can be applied to the copyright detection scene of video.Specifically, the feature of video is extracted, based on view
The feature of frequency and combine the mode of hash index determine whether there is with the duplicate video of the video, if it find that there is duplicate view
Frequently, and the upload side of video is not identical, then there may be invade video copy.
The embodiment of the present disclosure can also be applied to add tagged scene for video.Specifically, extracting the feature of video, base
Video tab relevant to the video is determined in the feature of video, and video tab is added to the video.
The embodiment of the present disclosure can also be applied to that video is avoided to repeat the scene uploaded.Specifically, if it is determined that in outlet
In the presence of content identical with video, then the upload procedure of video can be prevented, to economize on resources, and user be avoided to view repetition
Content.
In addition, the embodiment of the present disclosure can also be applied to video monitoring, target following, the detection of identification specific information, video
The several scenes such as teaser or tail detection.
Fig. 1 is shown can be using the video detecting method of the embodiment of the present disclosure or the exemplary system of video detecting device
The schematic diagram of framework.
As shown in Figure 1, system architecture 1000 may include one of terminal device 1001,1002,1003 or a variety of, net
Network 1004 and server 1005.Network 1004 is logical to provide between terminal device 1001,1002,1003 and server 1005
Believe the medium of link.Network 1004 may include various connection types, such as wired, wireless communication link or fiber optic cables etc.
Deng.
It should be understood that the number of terminal device, network and server in Fig. 1 is only schematical.According to realization need
It wants, can have any number of terminal device, network and server.For example server 1005 can be multiple server compositions
Server cluster etc..
User can be used terminal device 1001,1002,1003 and be interacted by network 1004 with server 1005, to receive
Or send message etc..Terminal device 1001,1002,1003 can be the various electronic equipments with display screen, including but unlimited
In smart phone, tablet computer, portable computer and desktop computer etc..
Server 1005 can be to provide the server of various services.For example, server 1005 can be from terminal device
1001, video to be detected is obtained on 1002,1003, key frame of video to be detected is extracted from video to be detected, is determined to be detected
The first global characteristics and the second global characteristics of key frame of video determine the second global characteristics with video frame sample is advanced with
Index out is compared, and is determined and indexed lower cryptographic Hash corresponding with video to be detected, and can be determined based on cryptographic Hash can
Can global characteristics relevant to video to be detected, the first global characteristics of video to be detected are compared with these global characteristics
It is right, to realize the detection to video to be detected.Some such as copyright detections can also be carried out for comparison result, video repeats to examine
Survey, the addition operation such as video tab, and feed back that these operate to terminal device 1001,1002,1003 as a result, to be accused to user
Know corresponding content.
For the process for extracting the first global characteristics and the second global characteristics, server 1005 can be closed video to be detected
The convolutional neural networks that key frame input one constructs in advance, the first full articulamentum and the second full articulamentum based on convolutional neural networks
Determine the first global characteristics and the second global characteristics.Furthermore it is possible to be instructed based on video frame sample to convolutional neural networks
Practice, to obtain above-mentioned index, and stores the video finger print library of the global characteristics of sample.
In addition, being directed to the training process of convolutional neural networks, can be executed by server 1005.It is, however, to be understood that
It is that other servers in addition to server 1005 or processing unit can also be utilized to execute training for convolutional neural networks
Journey, then, server 1005 can directly acquire the convolutional neural networks after training, can also realize the disclosure to video as a result,
Detection process.
It should be noted that the video detecting method that disclosure illustrative embodiments provide generally is held by server 1005
Row, correspondingly, video detecting device is generally disposed in server 1005.
However, following video detecting methods can also be realized by terminal device 1001,1002,1003, the disclosure to this not
It does specifically limited.
Fig. 2 shows the structures of the computer system of the electronic equipment suitable for being used to realize disclosure illustrative embodiments
Schematic diagram.
It should be noted that Fig. 2 shows the computer system 200 of electronic equipment be only an example, should not be to this public affairs
The function and use scope for opening embodiment bring any restrictions.
As shown in Fig. 2, computer system 200 includes central processing unit (CPU) 201, it can be read-only according to being stored in
Program in memory (ROM) 202 or be loaded into the program in random access storage device (RAM) 203 from storage section 208 and
Execute various movements appropriate and processing.In RAM203, it is also stored with various programs and data needed for system operatio.
CPU201, ROM202 and RAM203 are connected with each other by bus 204.Input/output (I/O) interface 205 is also connected to bus
204。
I/O interface 205 is connected to lower component: the importation 206 including keyboard, mouse etc.;It is penetrated including such as cathode
The output par, c 207 of spool (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.;Storage section 208 including hard disk etc.;
And the communications portion 209 of the network interface card including LAN card, modem etc..Communications portion 209 via such as because
The network of spy's net executes communication process.Driver 210 is also connected to I/O interface 205 as needed.Detachable media 211, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 210, in order to read from thereon
Computer program be mounted into storage section 208 as needed.
Particularly, in accordance with an embodiment of the present disclosure, it may be implemented as computer below with reference to the process of flow chart description
Software program.For example, embodiment of the disclosure includes a kind of computer program product comprising be carried on computer-readable medium
On computer program, which includes the program code for method shown in execution flow chart.In such reality
It applies in example, which can be downloaded and installed from network by communications portion 209, and/or from detachable media
211 are mounted.When the computer program is executed by central processing unit (CPU) 201, executes and limited in the system of the application
Various functions.
It should be noted that computer-readable medium shown in the disclosure can be computer-readable signal media or meter
Calculation machine readable storage medium storing program for executing either the two any combination.Computer readable storage medium for example can be --- but not
Be limited to --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor system, device or device, or any above combination.Meter
The more specific example of calculation machine readable storage medium storing program for executing can include but is not limited to: have the electrical connection, just of one or more conducting wires
Taking formula computer disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable type may be programmed read-only storage
Device (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device,
Or above-mentioned any appropriate combination.In the disclosure, computer readable storage medium can be it is any include or storage journey
The tangible medium of sequence, the program can be commanded execution system, device or device use or in connection.And at this
In open, computer-readable signal media may include in a base band or as the data-signal that carrier wave a part is propagated,
Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including but unlimited
In electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be that computer can
Any computer-readable medium other than storage medium is read, which can send, propagates or transmit and be used for
By the use of instruction execution system, device or device or program in connection.Include on computer-readable medium
Program code can transmit with any suitable medium, including but not limited to: wireless, electric wire, optical cable, RF etc. are above-mentioned
Any appropriate combination.
Flow chart and block diagram in attached drawing are illustrated according to the system of the various embodiments of the disclosure, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of above-mentioned module, program segment or code include one or more
Executable instruction for implementing the specified logical function.It should also be noted that in some implementations as replacements, institute in box
The function of mark can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are practical
On can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it wants
It is noted that the combination of each box in block diagram or flow chart and the box in block diagram or flow chart, can use and execute rule
The dedicated hardware based systems of fixed functions or operations is realized, or can use the group of specialized hardware and computer instruction
It closes to realize.
Being described in unit involved in the embodiment of the present disclosure can be realized by way of software, can also be by hard
The mode of part realizes that described unit also can be set in the processor.Wherein, the title of these units is in certain situation
Under do not constitute restriction to the unit itself.
As on the other hand, present invention also provides a kind of computer-readable medium, which be can be
Included in electronic equipment described in above-described embodiment;It is also possible to individualism, and without in the supplying electronic equipment.
Above-mentioned computer-readable medium carries one or more program, when the electronics is set by one for said one or multiple programs
When standby execution, so that method described in electronic equipment realization as the following examples.
Fig. 3 diagrammatically illustrates the flow chart of the video detecting method according to disclosure illustrative embodiments.With reference to figure
3, video detecting method may comprise steps of:
S32. video frame sample is obtained in advance, determines the first global characteristics and the second global characteristics of video frame sample, it will
Second global characteristics are converted to the first cryptographic Hash, are established and are indexed using the first cryptographic Hash, and the first global characteristics are stored in rope
Under drawing.
In the illustrative embodiments of the disclosure, video frame can be regarded as video it is static when the image that presents.Video
Frame sample can be the frame picture (or being single-frame images) extracted from existing video, it being understood, however, that, be not
The single image extracted from video can also be chosen for the video frame sample of the disclosure.
The global characteristics of video frame can indicate the Global Information of video frame images, be the low-level image feature of Pixel-level, can be with
It is related to the color characteristic of video frame images entirety, the textural characteristics of image entirety, spatial distribution characteristics of image entirety etc..It is global
Feature can be polymerize by the local feature for expressing image local information to be generated, and can also directly be extracted from the pixel layer of image
It generates.
First global characteristics and the second global characteristics described in the disclosure indicate the Global Information of corresponding video frame images.
When global characteristics to be expressed as to the vector of reflection image information, the dimension of the dimension of the first global characteristics and the second global characteristics
It is different.For example, the second global characteristics can be the image for further extracting feature from the first global characteristics and determining
Global Information, in this case, the dimension of the first global characteristics may may be for 512 dimensions, the dimension of the second global characteristics
32 dimensions.
The disclosure can use histograms of oriented gradients (Histogram of Oriented Gradient, HOG), scale
The methods of invariant features transformation (Scale-invariant Feature Transform, SIFT), deep learning network mention in advance
Take the global characteristics of video frame sample.
To be with the convolutional neural networks (Convolutional Neural Networks, CNN) in deep learning below
Example is illustrated the process of the first global characteristics and the second global characteristics that extract video frame sample.
The video frame sample obtained in advance can be input to convolutional neural networks by server, wherein convolutional neural networks
In include the first full articulamentum and the second full articulamentum.Wherein, in convolutional neural networks, the purpose of full articulamentum be in order to
The characteristic information by convolution and Chi Huahou is maximally utilized, these characteristic informations extracted before full articulamentum are gathered
It closes, to obtain the high-rise meaning of image, facilitates subsequent classification process that may be present.
Comprehensively consider model treatment speed and accuracy, the dimension of the first full articulamentum can in disclosure convolutional neural networks
Think 512 dimensions, the dimension of the second full articulamentum can be 32 dimensions.However, design of the invention, example also may be implemented in other dimensions
Such as, the dimension of the first full articulamentum can be 1024 dimensions, and the dimension of the second full articulamentum can be 64 dimensions.The disclosure does not do this
Limitation.
Disclosure illustrative embodiments using the feature determined via the first full articulamentum as the first global characteristics, and
Using the feature determined via the second full articulamentum as the second global characteristics.
The training process of convolutional neural networks is illustrated below.
With reference to Fig. 4, exemplary embodiment of the invention uses Triplet convolutional neural networks (Triplet CNN, ternary
Group CNN) network structure realizes the training process of network.It include three convolutional Neurals in Triplet convolutional neural networks structure
Network, and these three convolutional neural networks parameter sharings.It will be readily appreciated by those skilled in the art that when training, three convolution
The training sample of neural network input is different.
For the training process of Triplet convolutional neural networks, firstly, it is necessary to training sample is obtained, the example of the disclosure
Property embodiment can be using video frame sample as the training sample of network.Wherein, every group of video frame sample may include positive sample
This to and negative sample.
Server can determine positive sample pair, wherein positive sample is remembered respectively to may include identical two images of classification
For the first image and the second image, for example, can be using the first image as positive sample, using the second image as anchor point.
According to one embodiment of the disclosure, can use video often has the characteristics that space-time expending, obtains first
Image and the second image.Specifically, it is possible, firstly, to obtaining an image from existing video as the first image;Next, can
Using by video away from the image in the first image preset time period as the second image, wherein preset time period can be one section
The extremely short time, for example, 0.5s, so that the first image is identical as the classification of the second image.
With reference to Fig. 5, two images can be determined respectively as the first image and the second image in video streaming.Although figure
Shown in the first image and the second image be only separated by time of a video frame, however, it is understood that the first image with
Second image can also be separated by the time of multiple video frames.
In addition, can also determine the second figure using frame number in addition to determining the second image by means of preset time period
Picture.For example, the image of second frame after the first image in video can be extracted as the second figure after determining the first image
Picture does not do particular determination to this in this illustrative embodiment.
According to another embodiment of the present disclosure, it is contemplated that some personnel carry out form conversion to image, to escape copyright
This behavior can be described as copy-attack by the behavior of problem.First image can be carried out form conversion by server, with determination
Second image out.Wherein, form conversion can include but is not limited to posture conversion, scale conversion, modality conversion, illumination convert,
Colour temperature conversion etc., it can be seen that, the image converted through form described in the disclosure, it is only shape that substantial variation, which does not occur, for content
Formula is changed, thus, it is possible to think, image be belong to it is same category of.
Fig. 6 diagrammatically illustrates several examples of the image via form conversion.It is easily understood that except shown in Fig. 6 this
Outside several transformation results, those skilled in the art can also associate other transformation results, however, these should belong to the present invention
Design.
According to another embodiment of the present disclosure, both above situation can be integrated to determine the first image and the second figure
Picture.Specifically, firstly, server can obtain an image as the first candidate image from video, by video away from first
Image in candidate image preset time period is as the second candidate image;Next, server can be respectively to the first candidate figure
Picture and the second candidate image carry out form conversion, to determine candidate image collection, so it is easy to understand that candidate image is concentrated all
Image belong to same category;Then, server can be concentrated from candidate image and choose two images, respectively as the first figure
Picture and the second image, wherein the process specifically chosen can be realized by way of randomly selecting.
After determining the first image and the second image, that is, positive sample is being determined to rear, server can determine negative sample
This, the third image which can be different from the classification of positive sample pair.That is, can choose and the first image or
The different third image of two image scenes is as negative sample.Furthermore it is also possible to first determine negative sample, then determine positive sample
Right, the disclosure does not do particular determination to the sequence for determining an image in sample.
Determine positive sample to and in the case where negative sample, that is to say, that determining positive sample, anchor point and negative sample
In the case where, it can be using them as one group of training sample, as shown in figure 4, being input in convolutional neural networks, to realize once
Model training process.
It is easily understood that the process of above-mentioned determining training sample can be used, multiple groups training sample is determined, constantly to instruct
Practice convolutional neural networks.
In the training process of network, loss function (Loss Function) is needed to restrain.Loss function is also known as cost
Function (Cost Function) is that chance event or its value in relation to stochastic variable are mapped as nonnegative real number to indicate this
The risk of chance event or the function of loss.During being trained using sample to model, according to prediction probability and reality
Error between the probability of border is adjusted the parameter in model, constantly repeatedly inputs the process of sample, adjusting parameter as a result,
Until loss function reaches minimum, that is to say, that loss function tends towards stability, in this case, it is believed that loss function is received
It holds back, model training can terminate.
The loss function that the embodiment of the present disclosure uses may include range loss function, quantization loss function and comentropy damage
Lose function.Specifically, the weighted sum of range loss function, quantization loss function, comentropy loss function can be determined as rolling up
The loss function of product neural network.
Loss function L used by the embodiment of the present disclosure can be built as the form of formula 1:
L=α Lt+βLq+λLe(formula 1)
Wherein, LtIndicate range loss function, LqIndicate quantization loss function, LeIndicate comentropy loss function, α, β, λ
Respectively indicate range loss function Lt, quantify loss function LqWith comentropy loss function LeWeight (coefficient).
Range loss function LtIt can be built as the form of formula 2:
Lt=max { (xa-xp)+m-(xa-xn), 0 } (formula 2)
Wherein, xa、xp、xnThe feature of anchor point, positive sample and negative sample is respectively indicated, m indicates the confidence level of feature spacing,
It can be float.
Quantify loss function LqIt can be built as the form of formula 3:
Lq=| | | x(i)|-1||1(formula 3)
Wherein, x(i)Indicate any feature of training sample.
Comentropy loss function LeIt can be built as the form of formula 4:
Wherein, udIndicate the mean value of the feature of training sample, specifically, in the two-dimensional array formed by feature, characterization
The mean value of d column bit.
It should be understood that constructing the exemplary description that loss function is only the disclosure using above-mentioned formula 1.At it
In his embodiment, loss function used in network training can also be formed only with comentropy loss function shown in formula 4.
For convolutional neural networks structure shown in Fig. 4, those skilled in the art, so it is easy to understand that connect entirely except first
It connects except layer and the second full articulamentum, convolutional neural networks structure further includes various convolutional layers and pond layer.For example, convolutional Neural
Network structure can also include n Inception module, wherein consider model calculating speed and accuracy, n herein can
Think 5.Inception module can be considered as steroids convolutional layer, can provide different receptive fields, and it is each can to export capture
ReLU activation function can be used in the Feature Mapping of kind scale complex mode, convolutional layer.The specific network structure of CNN can be by opening
Hair personnel voluntarily adjust, and there is no special restriction on this for the disclosure.
Thus, it is possible to using the video frame sample determined, based on the loss function constructed to convolutional neural networks into
Row training, to determine the convolutional neural networks after training.It should be noted that the convolutional neural networks after practical application training
When, as shown in figure 4, three convolutional neural networks parameter sharings, therefore, when practical application network is to extract global characteristics, only
The concrete processing procedure of image is realized using a convolutional neural networks.
It should be understood that above-mentioned network training process can be executed by the server of execution video detecting method.However,
Convolutional neural networks after training can also be directly obtained by other processing unit training patterns, server.
It, can the first full articulamentum based on convolutional neural networks and second complete in the training process of convolutional neural networks
Articulamentum determines the first global characteristics and the second global characteristics of video frame sample respectively.That is, if complete by first
The feature that articulamentum determines is known as the first full articulamentum feature, and the feature that the second full articulamentum determines is known as the second full articulamentum
First full articulamentum feature then can be determined as above-mentioned first global characteristics, the second full articulamentum feature is determined as by feature
Above-mentioned second global characteristics.In addition, it should be noted that the first global characteristics and the second global characteristics are floating point type.
Server can carry out Hash processing to the second global characteristics of video frame sample, to determine the of video frame sample
The corresponding cryptographic Hash of two global characteristics, is denoted as the first cryptographic Hash.Wherein, Hash processing is called hashing, is by random length
Information MAP be shorter regular length binary value, this shorter binary value is known as cryptographic Hash.Then, server
It can use the first cryptographic Hash building index for determining video frame sample, and the first global characteristics of video frame sample stored
Under the index.
For example, being 000010 according to the cryptographic Hash that the second global characteristics of sample A are determined, 000010 building point is utilized
Bucket, the first global characteristics of sample A are floating point features, can store 000010 divide bucket under.If the Kazakhstan that sample B is determined
It is uncommon value also be 000010, then can by the first global characteristics of sample B there is also 000010 divide bucket under.It forms as a result,
Index corresponding relationship.
It is easily understood that be directed to each index value, can correspond to a point of bucket, each divide bucket be stored with one or
The floating point features of multiple samples.That is, each index value, which can correspond to one, carries out global characteristics ratio for detection-phase
Pair feature set.
S34. the key frame of video to be detected for extracting video to be detected determines that the first of key frame of video to be detected is global special
It seeks peace the second global characteristics, and the second global characteristics of key frame of video to be detected is converted into the second cryptographic Hash.
For the process for the key frame of video to be detected for extracting video to be detected, server can be every prefixed time interval
Video frame is extracted from video to be detected, as key frame of video to be detected, wherein prefixed time interval can taking human as determination,
It can also be extracted by fixed number of frames.That is, no matter the duration of video to be detected is how many, can therefrom extract
The key frame of video to be detected of fixed number of frames.For example, extracting 50 views from 1 minute video medium while spacing to be detected
Frequency frame is as key frame of video to be detected.
In addition, server can carry out analysis of complexity to frame image each in video to be detected, if complexity meets
Default complexity requirement, then can be using the frame image as key frame of video to be detected.Specifically, video frame images can be used
Gray level histogram second moment determine the complexity of each frame image.Special limit is not done in this illustrative embodiment to this
It is fixed.
After the key frame of video to be detected for extracting video to be detected, determine that the first of key frame of video to be detected is global special
It seeks peace the second global characteristics.Still for by the way of convolutional neural networks, key frame of video to be detected can be inputted
State training after convolutional neural networks in, and the first full articulamentum being based respectively in network and the second full articulamentum determine to
Detect the first global characteristics and the second global characteristics of key frame of video.
It is to be understood, however, that can also be determined using the methods of histograms of oriented gradients, Scale invariant features transform
The first global characteristics and the second global characteristics of key frame of video to be detected.
Next, server can carry out Hash processing to the second global characteristics of key frame of video to be detected, to obtain
The corresponding cryptographic Hash of the second global characteristics of key frame of video to be detected, as the second cryptographic Hash.
S36. when the similarity of the second cryptographic Hash and the first cryptographic Hash is greater than the first similarity threshold, the second Hash is extracted
It is worth corresponding first global characteristics, by the overall situation of corresponding first global characteristics of the second cryptographic Hash storage corresponding with the first cryptographic Hash
Feature is compared, and the testing result of video to be detected is determined according to comparison result.
After determining corresponding second cryptographic Hash of key frame of video to be detected, server can calculate the second cryptographic Hash with
The similarity of the first cryptographic Hash determined in step S32, if the similarity of the second cryptographic Hash and the first cryptographic Hash is greater than first
Similarity threshold, it is determined that the global characteristics of the corresponding storage of the first cryptographic Hash.It should be noted that corresponding storage described herein
It refers to that the first global characteristics of video frame sample are stored in the storing process under index in step S32.That is, corresponding
The global characteristics of storage are the first global characteristics of video frame sample.
For the process for calculating similarity, the Hamming distance of the second cryptographic Hash Yu the first cryptographic Hash, Hamming distance can be calculated
From smaller, similarity is bigger, in such a case, it is possible to determine whether the similarity of the second cryptographic Hash and the first cryptographic Hash is greater than
First similarity threshold.Wherein, the first similarity threshold can artificially be set in advance, and it is special that the disclosure does not do its value
Limitation.However, the other embodiments of the disclosure can also determine the second cryptographic Hash and rope using COS distance, Euclidean distance etc.
The similarity of first cryptographic Hash does not do particular determination to this in this illustrative embodiment.
After the global characteristics for determining the corresponding storage of the first cryptographic Hash, server can be by the second cryptographic Hash corresponding first
Global characteristics are compared with the global characteristics of the corresponding storage of the first cryptographic Hash.
In the illustrative embodiments of the disclosure, for the comparison of feature, server can calculate the second cryptographic Hash pair
The similarity of the global characteristics of the first global characteristics answered storage corresponding with the first cryptographic Hash, and similarity result is determined as comparing
To result.For example, the Euclidean distance between them can be calculated, and similarity is characterized with calculated Euclidean distance.However, also
Method can be determined using other similarities in addition to Euclidean distance, do not do special limit to this in this illustrative embodiment
It is fixed.
Server can be according to the overall situation of corresponding first global characteristics of the second cryptographic Hash storage corresponding with the first cryptographic Hash
The similarity of feature determines the testing result of video to be detected.Specifically, can determine whether video to be detected is existing
In existing video.
In addition, according to the global characteristics of corresponding first global characteristics of the second cryptographic Hash storage corresponding with the first cryptographic Hash
Similarity can execute the operation of corresponding scene.
According to some embodiments of the present disclosure, the video detecting method of disclosure illustrative embodiments can be applied to version
Power detection scene.Specifically, if the overall situation of corresponding first global characteristics of the second cryptographic Hash storage corresponding with the first cryptographic Hash
The similarity of feature is greater than the second similarity threshold, it is determined that the upload side of video to be detected identifies and the first cryptographic Hash is corresponding
The upload side of video identifies.
If upload side's mark of video to be detected is different from the upload side of the corresponding video of the first cryptographic Hash mark, may
There are same videos, there are two upload sides.In this case, server can be to two upload sides, video version
It weighs the side in regulatory agency or sends a warning message in many ways.
According to another embodiment of the present disclosure, the video detecting method of disclosure illustrative embodiments can be applied to
Label scene is added for video.Specifically, being deposited if corresponding first global characteristics of the second cryptographic Hash are corresponding with the first cryptographic Hash
The similarity of the global characteristics of storage is greater than third similarity threshold, it is determined that the first cryptographic Hash corresponds to the label of video.Wherein, should
Label can be added on existing video in advance by artificially, or be added on existing video in advance by the way of machine learning
Add.Next, the label that the first cryptographic Hash corresponds to video can be added to video to be detected.For example, video to be detected is one
The TV play name can be referred to as view to be detected in the case where corresponding video tab is TV play title by episodes
The label of frequency.Effect with reference to Fig. 7, after showing addition label.
According to another embodiment of the present disclosure, the video detecting method of disclosure illustrative embodiments can be applied to
It avoids repeating to upload scene.Specifically, if corresponding first global characteristics of the second cryptographic Hash storage corresponding with the first cryptographic Hash
Global characteristics similarity less than the 4th similarity threshold, then video to be detected is uploaded to corresponding platform.If it is greater than etc.
In the 4th similarity threshold, then video to be detected is prevented to upload, and send prompt information to the upload side of video to be detected, to accuse
Know that upload side's video content of video to be detected repeats.Thus, it is possible to which video is avoided to repeat to upload, resource has been saved.
It should be noted that above-mentioned first similarity threshold, the second similarity threshold, third similarity threshold and the 4th phase
It is respectively applied to different similarities like degree threshold value and judges scene, therefore, between them there is no incidence relation, can be difference and take
The threshold value of value.
Fig. 8 is shown under bonding wire according to an exemplary embodiment of the present disclosure and scheme realizes video detection on line
Schematic diagram.
For (offline) stage under line, global characteristics can be extracted from video frame sample using convolutional neural networks,
In, the global characteristics of extraction include the first global characteristics and the second global characteristics.On the one hand, Hash is carried out to the second global characteristics
Processing, obtains the first cryptographic Hash, is constructed and is indexed using the first cryptographic Hash;On the other hand, corresponding first global characteristics are stored
In video fingerprint data library, so that the first cryptographic Hash based on the second global characteristics can find the first global characteristics.
For the stage on line, the key frame of video to be detected of video to be detected is extracted, using convolutional neural networks to be checked
It surveys in key frame of video and extracts the first global characteristics and the second global characteristics.It can use the second global characteristics and based on structure under line
The index built determines the global characteristics of existing video corresponding with key frame of video to be detected, by key frame of video to be detected
First global characteristics are compared with corresponding global characteristics in video fingerprint data library, with the detection knot of determination video to be detected
Fruit.
It should be noted that although describing each step of method in the disclosure in the accompanying drawings with particular order, this is simultaneously
Undesired or hint must execute these steps in this particular order, or have to carry out the ability of step shown in whole
Realize desired result.Additional or alternative, it is convenient to omit multiple steps are merged into a step and executed by certain steps,
And/or a step is decomposed into execution of multiple steps etc..
Further, a kind of video detecting device is additionally provided in this example embodiment.
Fig. 9 diagrammatically illustrates the block diagram of the video detecting device of the illustrative embodiments of the disclosure.With reference to Fig. 9,
Video detecting device 9 according to an exemplary embodiment of the present disclosure may include detection preparation module 91, characteristic extracting module
93 and feature comparison module 95.
Specifically, detection preparation module 91 can be used for obtaining video frame sample in advance, the first of video frame sample is determined
Second global characteristics are converted to the first cryptographic Hash by global characteristics and the second global characteristics, are established and are indexed using the first cryptographic Hash,
And the first global characteristics are stored under index;Characteristic extracting module 93 can be used for extracting the video to be detected of video to be detected
Key frame, determines the first global characteristics and the second global characteristics of key frame of video to be detected, and by key frame of video to be detected
The second global characteristics be converted to the second cryptographic Hash;Feature comparison module 95 can be used for when the second cryptographic Hash and the first cryptographic Hash
Similarity when being greater than the first similarity threshold, corresponding first global characteristics of the second cryptographic Hash are extracted, by the second cryptographic Hash pair
The first global characteristics answered are compared with the global characteristics of the corresponding storage of the first cryptographic Hash, and are determined according to comparison result to be checked
Survey the testing result of video.
It can quickly be detected using the video detecting device of disclosure illustrative embodiments by providing index function
Out may global characteristics relevant to video to be detected, then using video to be detected itself global characteristics with may it is relevant entirely
Office's feature is compared, and can accurately determine out and whether there is video similar with video to be detected.
According to an exemplary embodiment of the present disclosure, using convolutional neural networks determine video frame the first global characteristics and
Second global characteristics, wherein global characteristics are full articulamentum feature.
Specifically, detection preparation module 91 can be used for obtaining video frame sample in advance, the video frame sample is determined
Described second full articulamentum Feature Conversion is the first cryptographic Hash, benefit by the first full articulamentum feature and the second full articulamentum feature
It is established and is indexed with first cryptographic Hash, and the floating point features of the described first full articulamentum is stored under the index;Feature
Extraction module 93 can be used for extracting the key frame of video to be detected of video to be detected, determine the key frame of video to be detected
First full articulamentum floating point features and the second full articulamentum feature, and by the second full articulamentum of the key frame of video to be detected
Feature Conversion is the second cryptographic Hash;Feature comparison module 95 can be used for when second cryptographic Hash and first cryptographic Hash
When similarity is greater than the first similarity threshold, extract the corresponding first full articulamentum floating point features of second cryptographic Hash with it is described
The full articulamentum floating point features of the corresponding storage of first cryptographic Hash is compared, and determines the video to be detected according to comparison result
Testing result.
The first global characteristics and the second global characteristics are extracted using convolutional neural networks, using the place of convolutional neural networks
Reason method, so that the video detection process versatility of disclosure illustrative embodiments is higher, it is easy to promote and utilize.
According to an exemplary embodiment of the present disclosure, video frame sample include positive sample to and negative sample;With reference to Figure 10, detection
Preparation module 91 further includes sample acquisition unit 101.
Specifically, sample acquisition unit 101 can be configured as execution: obtaining positive sample pair in advance, positive sample is to including
First image and the second image;Wherein, the first image is identical as the classification of the second image;Classification and positive sample pair are obtained in advance
The different third image of classification, as negative sample;Wherein, using the positive sample to and negative sample to the convolutional neural networks
It is trained, to adjust the parameter in the convolutional neural networks.
Using network is trained based on the Triplet CNN network structure of positive negative sample, Triplet CNN network structure is made
For a kind of twin neural network (Siamese Convolution Neural Network, SCNN), when determining similarity, more
It is accurate to add.
According to an exemplary embodiment of the present disclosure, sample acquisition unit 101 can be additionally configured to execute: from video sample
One image of middle acquisition, as the first image;Using in video sample away from the image in the first image preset time period as second
Image.
According to an exemplary embodiment of the present disclosure, sample acquisition unit 101 can be additionally configured to execute: from video sample
One image of middle acquisition, as the first image;Form conversion is carried out to the first image, to determine the second image.
According to an exemplary embodiment of the present disclosure, sample acquisition unit 101 can be additionally configured to execute: from video sample
One image of middle acquisition, as the first candidate image;By in video sample away from the image in the first candidate image preset time period
As the second candidate image;Form conversion is carried out to the first candidate image and the second candidate image respectively, to determine candidate image
Collection;It is concentrated from candidate image and obtains two images, respectively as the first image and the second image.
According to an exemplary embodiment of the present disclosure, with reference to Figure 11, characteristic extracting module 93 may include network training unit
1101 and feature extraction unit 1103.
Specifically, network training unit 1101 can be used for video frame sample inputting convolutional neural networks, convolution is calculated
The loss function of neural network restrains the loss function of convolutional neural networks, to determine the convolutional neural networks after training;
Feature extraction unit 1103 can be used for the convolutional Neural net after the key frame of video to be detected input training by video to be detected
Network, using the feature determined via the first full articulamentum as the first global characteristics of key frame of video to be detected, and will be via
Second global characteristics of the feature that second full articulamentum is determined as key frame of video to be detected.
According to an exemplary embodiment of the present disclosure, the loss function of convolutional neural networks includes range loss function, quantization
Loss function and comentropy loss function;
Range loss function are as follows: Lt=max { (xa-xp)+m-(xa-xn),0};
Quantify loss function are as follows: Lq=| | | x(i)|-1||1;
Comentropy loss function are as follows:
Wherein, LtIndicate range loss function, xa、xp、xnRespectively indicate the spy of two samples and negative sample of positive sample centering
Sign, m indicate the confidence level of feature spacing;LqIndicate quantization loss function, x(i)Indicate any feature of video frame sample;LeIt indicates
Comentropy loss function, udIndicate the mean value of the feature of video frame sample.
According to an exemplary embodiment of the present disclosure, with reference to Figure 12, characteristic extracting module 93 can also include key-frame extraction
Unit 1201.
Specifically, key-frame extraction unit 1201 can be used for extracting view from video to be detected every prefixed time interval
Frequency frame, as key frame of video to be detected.
According to an exemplary embodiment of the present disclosure, with reference to Figure 13, video detecting device 13 compared to video detecting device 9,
It can also include copyright detection module 131.
Specifically, copyright detection module 131 can be configured as execution: if second cryptographic Hash corresponding first is entirely
The similarity of office's feature global characteristics corresponding with first cryptographic Hash is greater than the second similarity threshold, it is determined that described to be checked
The uploads side for surveying video identifies and first cryptographic Hash corresponds to the upload side of video and identifies, upper when the video to be detected
When the upload side that biography side's mark and first cryptographic Hash correspond to video identifies different, send a warning message.
By the video detection process of the present exemplary embodiment, video copy can detecte.
According to an exemplary embodiment of the present disclosure, with reference to Figure 14, video detecting device 14 compared to video detecting device 9,
It can also include label adding module 141.
Specifically, label adding module 141 can be configured as execution: if second cryptographic Hash corresponding first is entirely
The similarity of office's feature global characteristics corresponding with first cryptographic Hash is greater than third similarity threshold, it is determined that described first
Cryptographic Hash corresponds to the label of video, and the label that first cryptographic Hash corresponds to video is added to the video to be detected.
By the video detection process of the present exemplary embodiment, label can be added for video clip.
According to an exemplary embodiment of the present disclosure, with reference to Figure 15, video detecting device 15 compared to video detecting device 9,
It can also include video uploading module 151.
If specifically, video upload device 151 can be used for corresponding first global characteristics of second cryptographic Hash with
The similarity of the corresponding global characteristics of first cryptographic Hash, then will be on the video to be detected less than the 4th similarity threshold
It passes.
By the video detection process of the present exemplary embodiment, can repeat to upload to avoid video.
Since each functional module and the above method of the program analysis of running performance device of embodiment of the present invention are invented
It is identical in embodiment, therefore details are not described herein.
Through the above description of the embodiments, those skilled in the art is it can be readily appreciated that example described herein is implemented
Mode can also be realized by software realization in such a way that software is in conjunction with necessary hardware.Therefore, according to the disclosure
The technical solution of embodiment can be embodied in the form of software products, which can store non-volatile at one
Property storage medium (can be CD-ROM, USB flash disk, mobile hard disk etc.) in or network on, including some instructions are so that a calculating
Equipment (can be personal computer, server, terminal installation or network equipment etc.) is executed according to disclosure embodiment
Method.
In addition, above-mentioned attached drawing is only the schematic theory of processing included by method according to an exemplary embodiment of the present invention
It is bright, rather than limit purpose.It can be readily appreciated that the time that above-mentioned processing shown in the drawings did not indicated or limited these processing is suitable
Sequence.In addition, be also easy to understand, these processing, which can be, for example either synchronously or asynchronously to be executed in multiple modules.
It should be noted that although being referred to several modules or list for acting the equipment executed in the above detailed description
Member, but this division is not enforceable.In fact, according to embodiment of the present disclosure, it is above-described two or more
Module or the feature and function of unit can embody in a module or unit.Conversely, an above-described mould
The feature and function of block or unit can be to be embodied by multiple modules or unit with further division.
Those skilled in the art after considering the specification and implementing the invention disclosed here, will readily occur to its of the disclosure
His embodiment.This application is intended to cover any variations, uses, or adaptations of the disclosure, these modifications, purposes or
Adaptive change follow the general principles of this disclosure and including the undocumented common knowledge in the art of the disclosure or
Conventional techniques.The description and examples are only to be considered as illustrative, and the true scope and spirit of the disclosure are by claim
It points out.
It should be understood that the present disclosure is not limited to the precise structures that have been described above and shown in the drawings, and
And various modifications and changes may be made without departing from the scope thereof.The scope of the present disclosure is only limited by the attached claims.
Claims (13)
1. a kind of video detecting method characterized by comprising
Video frame sample is obtained in advance, determines the first global characteristics and the second global characteristics of the video frame sample, it will be described
Second global characteristics are converted to the first cryptographic Hash, are established and are indexed using first cryptographic Hash, and by first global characteristics
It is stored under the index;
The key frame of video to be detected for extracting video to be detected, determine the key frame of video to be detected the first global characteristics and
Second global characteristics, and the second global characteristics of the key frame of video to be detected are converted into the second cryptographic Hash;
When the similarity of second cryptographic Hash and first cryptographic Hash is greater than the first similarity threshold, described second is extracted
Corresponding first global characteristics of cryptographic Hash, by corresponding first global characteristics of second cryptographic Hash and first cryptographic Hash pair
The global characteristics that should be stored are compared, and the testing result of the video to be detected is determined according to comparison result.
2. video detecting method according to claim 1, which is characterized in that use convolution mind to the processing of the video frame
Through network, the global characteristics are full articulamentum feature;The method specifically includes:
Video frame sample is obtained in advance, determines that the first full articulamentum feature of the video frame sample and the second full articulamentum are special
Described second full articulamentum Feature Conversion is the first cryptographic Hash by sign, establishes index using first cryptographic Hash, and will be described
The floating point features of first full articulamentum is stored under the index;
The key frame of video to be detected for extracting video to be detected determines that the first full articulamentum of the key frame of video to be detected is floating
Point feature and the second full articulamentum feature, and be second by the second of the key frame of video to be detected the full articulamentum Feature Conversion
Cryptographic Hash;
When the similarity of second cryptographic Hash and first cryptographic Hash is greater than the first similarity threshold, described second is extracted
The full articulamentum floating point features of the corresponding first full articulamentum floating point features of cryptographic Hash storage corresponding with first cryptographic Hash into
Row compares, and the testing result of the video to be detected is determined according to comparison result.
3. video detecting method according to claim 2, which is characterized in that the video frame sample include positive sample to
Negative sample;Wherein, obtaining video frame sample in advance includes:
Positive sample pair is obtained in advance, and the positive sample is to including the first image and the second image;Wherein, the first image and institute
The classification for stating the second image is identical;
The classification third image different from the classification of the positive sample pair is obtained in advance, as negative sample;
Wherein, using the positive sample to and negative sample the convolutional neural networks are trained, to adjust convolution mind
Through the parameter in network.
4. video detecting method according to claim 3, which is characterized in that obtain positive sample to including:
An image is obtained from video sample, as the first image;
Using in the video sample away from the image in the first image preset time period as the second image.
5. video detecting method according to claim 3, which is characterized in that obtain positive sample to including:
An image is obtained from video sample, as the first image;
Form conversion is carried out to the first image, to determine the second image.
6. video detecting method according to claim 3, which is characterized in that obtain positive sample to including:
An image is obtained from video sample, as the first candidate image;
Using in the video sample away from the image in the first candidate image preset time period as the second candidate image;
Form conversion is carried out to first candidate image and second candidate image respectively, to determine candidate image collection;
It is concentrated from the candidate image and obtains two images, respectively as the first image and the second image.
7. the video detecting method according to any one of claim 2 to 6, which is characterized in that determine the view to be detected
The first global characteristics and the second global characteristics of frequency key frame include:
The video frame sample is inputted into the convolutional neural networks, the loss function of the convolutional neural networks is calculated, makes institute
The loss function convergence of convolutional neural networks is stated, to determine the convolutional neural networks after training;
It, will be via described first by the convolutional neural networks after the key frame of video to be detected input training of the video to be detected
First global characteristics of the feature that full articulamentum is determined as the key frame of video to be detected, and will be complete via described second
Second global characteristics of the feature that articulamentum is determined as the key frame of video to be detected.
8. video detecting method according to claim 7, which is characterized in that the loss function packet of the convolutional neural networks
Include range loss function, quantization loss function and comentropy loss function;
The range loss function are as follows: Lt=max { (xa-xp)+m-(xa-xn),0};
The quantization loss function are as follows: Lq=| | | x(i)|-1||1;
The comentropy loss function are as follows:
Wherein, LtIndicate range loss function, xa、xp、xnRespectively indicate the feature of two samples and negative sample of positive sample centering, m
Indicate the confidence level of feature spacing;LqIndicate quantization loss function, x(i)Indicate any feature of video frame sample;LeIndicate information
Entropy loss function, udIndicate the mean value of the feature of video frame sample.
9. video detecting method according to claim 1, which is characterized in that the video to be detected for extracting video to be detected closes
Key frame includes:
Video frame is extracted from video to be detected every prefixed time interval, as key frame of video to be detected.
10. video detecting method according to claim 1, which is characterized in that the video detecting method further include:
If corresponding first global characteristics of second cryptographic Hash global characteristics corresponding with first cryptographic Hash is similar
Degree is greater than the second similarity threshold, it is determined that the upload side of the video to be detected identifies and the corresponding view of first cryptographic Hash
The upload side of frequency identifies, when the side of uploading of the video to be detected identifies and first cryptographic Hash corresponds to the upload side of video
When identifying different, send a warning message;Or
If corresponding first global characteristics of second cryptographic Hash global characteristics corresponding with first cryptographic Hash is similar
Degree is greater than third similarity threshold, it is determined that first cryptographic Hash corresponds to the label of video, and by first cryptographic Hash pair
The label of video is answered to be added to the video to be detected;Or
If corresponding first global characteristics of second cryptographic Hash global characteristics corresponding with first cryptographic Hash is similar
Degree then uploads the video to be detected less than the 4th similarity threshold.
11. a kind of video detecting device characterized by comprising
Preparation module is detected, for obtaining video frame sample in advance, determines the first global characteristics and the of the video frame sample
Second global characteristics are converted to the first cryptographic Hash by two global characteristics, are established and are indexed using first cryptographic Hash, and will
First global characteristics are stored under the index;
Characteristic extracting module determines the Video Key to be detected for extracting the key frame of video to be detected of video to be detected
The first global characteristics and the second global characteristics of frame, and the second global characteristics of the key frame of video to be detected are converted to
Two cryptographic Hash;
Feature comparison module is greater than the first similarity threshold for the similarity when second cryptographic Hash and first cryptographic Hash
When value, corresponding first global characteristics of second cryptographic Hash are extracted, by corresponding first global characteristics of second cryptographic Hash
It is compared with the global characteristics of the corresponding storage of first cryptographic Hash, and the video to be detected is determined according to comparison result
Testing result.
12. a kind of computer-readable medium, is stored thereon with computer program, which is characterized in that described program is held by processor
The video detecting method as described in any one of claims 1 to 10 is realized when row.
13. a kind of electronic equipment characterized by comprising
One or more processors;
Storage device, for storing one or more programs, when one or more of programs are by one or more of processing
When device executes, so that one or more of processors realize the video detection side as described in any one of claims 1 to 10
Method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910226798.4A CN110163079A (en) | 2019-03-25 | 2019-03-25 | Video detecting method and device, computer-readable medium and electronic equipment |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910226798.4A CN110163079A (en) | 2019-03-25 | 2019-03-25 | Video detecting method and device, computer-readable medium and electronic equipment |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110163079A true CN110163079A (en) | 2019-08-23 |
Family
ID=67638871
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910226798.4A Pending CN110163079A (en) | 2019-03-25 | 2019-03-25 | Video detecting method and device, computer-readable medium and electronic equipment |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110163079A (en) |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110996123A (en) * | 2019-12-18 | 2020-04-10 | 广州市百果园信息技术有限公司 | Video processing method, device, equipment and medium |
CN111182364A (en) * | 2019-12-27 | 2020-05-19 | 杭州趣维科技有限公司 | Short video copyright detection method and system |
CN111294610A (en) * | 2020-02-20 | 2020-06-16 | 北京奇艺世纪科技有限公司 | Video processing method and device |
CN111476868A (en) * | 2020-04-07 | 2020-07-31 | 哈尔滨工业大学 | Animation generation model training and animation generation method and device based on deep learning |
CN111507289A (en) * | 2020-04-22 | 2020-08-07 | 上海眼控科技股份有限公司 | Video matching method, computer device and storage medium |
CN111967303A (en) * | 2020-06-30 | 2020-11-20 | 西安电子科技大学 | Fingerprint template protection method based on maximum-minimum index hash |
CN112115286A (en) * | 2020-08-06 | 2020-12-22 | 国网安徽省电力有限公司电力科学研究院 | Robot environment identification method and system based on deep reinforcement learning |
CN112115295A (en) * | 2020-08-27 | 2020-12-22 | 广州华多网络科技有限公司 | Video image detection method and device and electronic equipment |
CN112215908A (en) * | 2020-10-12 | 2021-01-12 | 国家计算机网络与信息安全管理中心 | Compressed domain-oriented video content comparison system, optimization method and comparison method |
CN112381151A (en) * | 2020-11-17 | 2021-02-19 | 有米科技股份有限公司 | Similar video determination method and device |
CN112712090A (en) * | 2019-10-24 | 2021-04-27 | 北京易真学思教育科技有限公司 | Image processing method, device, equipment and storage medium |
CN113347082A (en) * | 2021-08-06 | 2021-09-03 | 深圳康易世佳科技有限公司 | Method and device for intelligently displaying shared messages containing short videos |
CN113559506A (en) * | 2021-09-24 | 2021-10-29 | 深圳易帆互动科技有限公司 | Automatic testing method and device for frame synchronization and readable storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101464909A (en) * | 2009-01-20 | 2009-06-24 | 清华大学 | Fast robust approximately same video detection and exclusion method |
CN101711392A (en) * | 2007-04-13 | 2010-05-19 | 艾法罗媒体有限责任公司 | Video detection system and methods |
US20120321181A1 (en) * | 2011-06-20 | 2012-12-20 | Microsoft Corporation | Near-duplicate video retrieval |
CN103336795A (en) * | 2013-06-09 | 2013-10-02 | 华中科技大学 | Video indexing method based on multiple features |
CN103617233A (en) * | 2013-11-26 | 2014-03-05 | 烟台中科网络技术研究所 | Method and device for detecting repeated video based on semantic content multilayer expression |
US20180307706A1 (en) * | 2015-12-30 | 2018-10-25 | Huawei Technologies Co., Ltd. | Image Query Method and Apparatus |
CN109241327A (en) * | 2017-07-03 | 2019-01-18 | 北大方正集团有限公司 | Image search method and device |
-
2019
- 2019-03-25 CN CN201910226798.4A patent/CN110163079A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101711392A (en) * | 2007-04-13 | 2010-05-19 | 艾法罗媒体有限责任公司 | Video detection system and methods |
CN101464909A (en) * | 2009-01-20 | 2009-06-24 | 清华大学 | Fast robust approximately same video detection and exclusion method |
US20120321181A1 (en) * | 2011-06-20 | 2012-12-20 | Microsoft Corporation | Near-duplicate video retrieval |
CN103336795A (en) * | 2013-06-09 | 2013-10-02 | 华中科技大学 | Video indexing method based on multiple features |
CN103617233A (en) * | 2013-11-26 | 2014-03-05 | 烟台中科网络技术研究所 | Method and device for detecting repeated video based on semantic content multilayer expression |
US20180307706A1 (en) * | 2015-12-30 | 2018-10-25 | Huawei Technologies Co., Ltd. | Image Query Method and Apparatus |
CN109241327A (en) * | 2017-07-03 | 2019-01-18 | 北大方正集团有限公司 | Image search method and device |
Non-Patent Citations (1)
Title |
---|
张乃光;石慧杰;祝晓斌;: "基于深度哈希的多模态视频拷贝检测方法", 广播电视信息, no. 1, 11 October 2018 (2018-10-11) * |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112712090A (en) * | 2019-10-24 | 2021-04-27 | 北京易真学思教育科技有限公司 | Image processing method, device, equipment and storage medium |
CN110996123A (en) * | 2019-12-18 | 2020-04-10 | 广州市百果园信息技术有限公司 | Video processing method, device, equipment and medium |
CN111182364A (en) * | 2019-12-27 | 2020-05-19 | 杭州趣维科技有限公司 | Short video copyright detection method and system |
CN111294610A (en) * | 2020-02-20 | 2020-06-16 | 北京奇艺世纪科技有限公司 | Video processing method and device |
CN111476868B (en) * | 2020-04-07 | 2023-06-23 | 哈尔滨工业大学 | Animation generation model training and animation generation method and device based on deep learning |
CN111476868A (en) * | 2020-04-07 | 2020-07-31 | 哈尔滨工业大学 | Animation generation model training and animation generation method and device based on deep learning |
CN111507289A (en) * | 2020-04-22 | 2020-08-07 | 上海眼控科技股份有限公司 | Video matching method, computer device and storage medium |
CN111967303A (en) * | 2020-06-30 | 2020-11-20 | 西安电子科技大学 | Fingerprint template protection method based on maximum-minimum index hash |
CN111967303B (en) * | 2020-06-30 | 2023-08-22 | 西安电子科技大学 | Fingerprint template protection method based on maximum-minimum index hash |
CN112115286A (en) * | 2020-08-06 | 2020-12-22 | 国网安徽省电力有限公司电力科学研究院 | Robot environment identification method and system based on deep reinforcement learning |
CN112115295A (en) * | 2020-08-27 | 2020-12-22 | 广州华多网络科技有限公司 | Video image detection method and device and electronic equipment |
CN112215908A (en) * | 2020-10-12 | 2021-01-12 | 国家计算机网络与信息安全管理中心 | Compressed domain-oriented video content comparison system, optimization method and comparison method |
CN112381151A (en) * | 2020-11-17 | 2021-02-19 | 有米科技股份有限公司 | Similar video determination method and device |
CN112381151B (en) * | 2020-11-17 | 2024-03-22 | 有米科技股份有限公司 | Method and device for determining similar videos |
CN113347082A (en) * | 2021-08-06 | 2021-09-03 | 深圳康易世佳科技有限公司 | Method and device for intelligently displaying shared messages containing short videos |
CN113559506A (en) * | 2021-09-24 | 2021-10-29 | 深圳易帆互动科技有限公司 | Automatic testing method and device for frame synchronization and readable storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110163079A (en) | Video detecting method and device, computer-readable medium and electronic equipment | |
CN108898086B (en) | Video image processing method and device, computer readable medium and electronic equipment | |
Wang et al. | SaliencyGAN: Deep learning semisupervised salient object detection in the fog of IoT | |
CN108197532B (en) | The method, apparatus and computer installation of recognition of face | |
CN108304835A (en) | character detecting method and device | |
CN108171191B (en) | Method and apparatus for detecting face | |
CN109117831A (en) | The training method and device of object detection network | |
US20120027252A1 (en) | Hand gesture detection | |
CN109446990A (en) | Method and apparatus for generating information | |
CN111275784B (en) | Method and device for generating image | |
TW202207077A (en) | Text area positioning method and device | |
CN109308490A (en) | Method and apparatus for generating information | |
CN108564102A (en) | Image clustering evaluation of result method and apparatus | |
CN113434716B (en) | Cross-modal information retrieval method and device | |
CN111881777B (en) | Video processing method and device | |
US11804043B2 (en) | Detecting objects in a video using attention models | |
CN110443824A (en) | Method and apparatus for generating information | |
CN109389096A (en) | Detection method and device | |
CN109871749A (en) | A kind of pedestrian based on depth Hash recognition methods and device, computer system again | |
CN110019939A (en) | Video temperature prediction technique, device, terminal device and medium | |
WO2023024413A1 (en) | Information matching method and apparatus, computer device and readable storage medium | |
CN109325539B (en) | Insulator string drop detection method and device | |
CN113515669A (en) | Data processing method based on artificial intelligence and related equipment | |
CN115131698A (en) | Video attribute determination method, device, equipment and storage medium | |
CN113762314A (en) | Smoke and fire detection method and device |
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 |