CN106469314A - A kind of video image classifier method based on space-time symbiosis binary-flow network - Google Patents

A kind of video image classifier method based on space-time symbiosis binary-flow network Download PDF

Info

Publication number
CN106469314A
CN106469314A CN201610794689.9A CN201610794689A CN106469314A CN 106469314 A CN106469314 A CN 106469314A CN 201610794689 A CN201610794689 A CN 201610794689A CN 106469314 A CN106469314 A CN 106469314A
Authority
CN
China
Prior art keywords
time
network
space
symbiosis
binary
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
Application number
CN201610794689.9A
Other languages
Chinese (zh)
Inventor
夏春秋
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Vision Technology Co Ltd
Original Assignee
Shenzhen Vision Technology Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Shenzhen Vision Technology Co Ltd filed Critical Shenzhen Vision Technology Co Ltd
Priority to CN201610794689.9A priority Critical patent/CN106469314A/en
Publication of CN106469314A publication Critical patent/CN106469314A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • 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
    • G06N3/08—Learning methods

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Image Analysis (AREA)

Abstract

A kind of method of the video image classifier based on space-time symbiosis binary-flow network proposing in the present invention, its main contents includes：Data input, space-time binary-flow network, fusion, SVM classifier, its process is, input picture and Optic flow information first, and binding time network and spatial network carry out early stage and merge, it is input to merging output among SVM classifier as characteristic vector, obtain final classification result.The present invention, using the early binary-flow network binding time melting and spatial information (space-time symbiosis) method, using the sets of video data of monkey class, uses more frames (i.e. more spatial datas) to produce a significant precision from each video and improves；The combination of room and time information, both form complementation, and precision reaches 65.8%.Formed by using the less detached cluster of the method for symbiosis, with detached cluster often more closely together with, better profit from temporal information.

Description

A kind of video image classifier method based on space-time symbiosis binary-flow network
Technical field
The present invention relates to video image classifier field, especially relate to a kind of video based on space-time symbiosis binary-flow network Image classification method.
Background technology
Video image classifier is a problem having very much challenge, because posture and cosmetic variation cause big internal type Change, the little interior change that also nuance in the overall appearance between type causes.Recently, depth convolution god It is used to learn many powerful functions through network (DCNNs), process big change with hierarchical model, be automatically positioned area Domain.Although these methods have improved, former work is asked the classification task of object as a static image classification Topic, ignores complementary temporal information present in video.So far, not using the method based on neural net method come to regarding The object of frequency is classified.
Present invention introduces the object classification problem based on video, employ the early binary-flow network binding time melting and space letter Breath (space-time symbiosis) method, first input picture and Optic flow information, binding time network and spatial network carry out merging in early days, will Merge output to be input among SVM classifier as characteristic vector, obtain final classification result.From each video using more Frame (i.e. more spatial datas) produces a significant precision and improves；The combination of room and time information, both form complementation, Precision reaches 65.8%.Merged using early stage is to have a potential defect, the information fusion of room and time because late period merges It is to finally complete, which has limited the quantity (or decision) of the side information obtaining from SoftMax classification layer combination, so, lead to Cross and formed using the less detached cluster of the method for symbiosis, with detached cluster often more closely together with, Ke Yigeng Good land productivity temporal information.
Content of the invention
For have ignored the problem that video data is classified, it is an object of the invention to provide a kind of be based on space-time altogether The video image classifier method of raw binary-flow network, using the early binary-flow network binding time melting and spatial information (space-time symbiosis) side Method, using the sets of video data of monkey class, improves classification performance.
For solving the above problems, the present invention provides a kind of video image classifier method based on space-time symbiosis binary-flow network, Its main contents includes：
(1) data input；
(2) space-time binary-flow network；
(3) merge；
(4) SVM classifier.
Wherein, a kind of video image classifier method based on space-time symbiosis binary-flow network, using the early binary-flow network melting knot Conjunction time and spatial information (space-time symbiosis) method, using the sets of video data of monkey class, use more frames (i.e. from each video More spatial datas) produce a significant precision raising；The combination of room and time information, both form complementation, precision Reach 65.8%.
Wherein, a kind of video image classifier method based on space-time symbiosis binary-flow network, the visualization technique based on dimensionality reduction Drawn using t- distribution neighborhood embedded mobile GIS (t-SNE), formed by using the less detached cluster of the method for symbiosis, and point From cluster often more closely together, better profit from temporal information.
Wherein, described data input, including image and Optic flow information, data set is made up of 100 kinds of monkey class video sets；Should Data set is divided into training set and test set.Record monkey class video in certain distance, this data set has larger challenge, such as big The camera motion change of scale and sizable attitudes vibration；
There is provided data below for each class (monkey kind)：There is the video clipping of active annotations, sound clip, surround Box, and classification and distributing position.
Further, described test, each video clipping is tested by the way of 5 frames (FPS) per second, calculates every 5 frames Light stream carry out computational efficiency.
Wherein, described space-time binary-flow network, including time network, spatial network, and space-time symbiosis decoding.
Further, described time network and spatial network, including
(1) sequential networkIt is used as level stream Ox, vertically flow OySize O with light streammagInput combine to form one Individual single optical signature mapping O ∈ Rh×w×3, wherein h × w is the size of Feature Mapping (image)；
(2) spatial networkRGB frame (image) is used as input；
WithAll using DCNN structure, form S by 5 convolutional layersc1,Sc2,…,Sc5, next to that being fully connected a layer Sfc6；Should Network is trained, and is a single example by the incoming frame (image or light stream) of each video, using pre-training net Network；When being classified, each image (or light stream of frame) is initially treated as independent；N for a videofFrame generates NfClassification determines.
Further, described space-time symbiosis decoding, including by the common united room and time feature occurring, makes Pass through to calculate the convolutional layer that symbiosis combines room and time network with DCNNs method, specifically, allow the n-th layer of time-space network Feature Mapping beWithdnIt is the number of the dimension of Feature Mapping, calculate Feature Mapping combination
WithRefer to the local feature vectors of the room and time stream on position (i, j), carry out vector Change operation,Symbiosis feature as position (i, j)；Therefore, in the symbiosis of each locus Pattern, the visual movement of apposition computing capture, maximum pond is applied to all of local coder vector；Pi,jCreate last feature RepresentFinally, L2Standardization is applied to coding vector；
Space-time bilinearity DCNN feature combines fc6 space-time characteristic and is used for double fluid early stage fusion, and this enables us to combine office Portion and the room and time information of the overall situation.
Wherein, described fusion, comprises the steps：
(1) early stage merges
(2) two independent sequential networks and spatial network S are usedoWithBinary-flow network be used for action recognition, early stage In conjunction with double-current information, by merging fc6 output, Sfc6And Tfc6, fc6 is first full articulamentum, is often used in from DCNNs Extract single feature；We are referred to as this network, and double fluid is early to be melted；
(3) carry out merging in early days, the method then merging double fluid using bilinearity DCNN, room and time information phase is tied Close；The combination initial data pretreatment of the layer by being fully connected, determine classified counting object, carry out point on calculating object Class.
Wherein, described SVM classifier, its principle is：
If linear separability sample set and be (xi,yi), i=1 ..., n, x ∈ Rd, y ∈ {+1, -1 } is category label, then
W x+b=0
It is the classifying face equation of SVM classifier；
In classification, in order that classifying face is correctly classified by all samples and class interval reaches maximum, under needing to meet Two, face condition：
Φ (x)=min (wTw)
yi(w·xi+b)-1≥0
Can be obtained by optimal classification surface by solving this constrained optimization problems, and cross nearest from classifying face in two class samples Put and be just so that, parallel to the training sample on the hyperplane of optimal classification surface, those special sample that in formula, equal sign is set up, Because they support optimal classification surface, therefore it is referred to as support vector；Fusion output is input to SVM as characteristic vector divide Among class device, obtain final classification result.
Brief description
Fig. 1 is a kind of system flow chart of the video image classifier method based on space-time symbiosis binary-flow network of the present invention.
Fig. 2 is a kind of video data of the monkey class of the video image classifier method based on space-time symbiosis binary-flow network of the present invention Collection.
Fig. 3 is that a kind of use T-SNE of the video image classifier method based on space-time symbiosis binary-flow network of the present invention is visual Change qualitative evaluation.
Fig. 4 is a kind of space-time symbiosis concept map of the video image classifier method based on space-time symbiosis binary-flow network of the present invention Method.
Fig. 5 is that a kind of early stage convergence strategy of the video image classifier method based on space-time symbiosis binary-flow network of the present invention is shown It is intended to.
Fig. 6 is a kind of monkey class example positioning of video image classifier method based on space-time symbiosis binary-flow network of the present invention Figure.
Specific embodiment
It should be noted that in the case of not conflicting, the embodiment in the application and the feature in embodiment can phases Mutually combine, with specific embodiment, the present invention is described in further detail below in conjunction with the accompanying drawings.
Fig. 1 is a kind of system flow chart of the video image classifier method based on space-time symbiosis binary-flow network of the present invention.Main Data input to be included；Space-time binary-flow network；Merge；SVM classifier.
Data input includes image and Optic flow information, and data set is made up of 100 kinds of monkey class video sets；This data set is divided into instruction Practice collection and test set.Record monkey class video in certain distance, this data set has larger challenge, such as large-scale camera fortune Dynamic change and sizable attitudes vibration；There is provided data below for each class (monkey kind)：The video with active annotations is cut Volume, sound clip, autonavigator detects, and classification and distributing position.
Wherein, each video clipping is tested by the way of 5 frames (FPS) per second, the light stream calculating every 5 frames is calculating effect Rate.
Space-time binary-flow network, including time network, spatial network, and space-time symbiosis decoding.
Wherein, time network and spatial network, including
(1) sequential networkIt is used as level stream Ox, vertically flow OySize O with light streammagInput combine to form one Individual single optical signature mapping O ∈ Rh×w×3, wherein h × w is the size of Feature Mapping (image)；
(2) spatial networkRGB frame (image) is used as input；
WithAll using DCNN structure, form S by 5 convolutional layersc1,Sc2,…,Sc5, next to that being fully connected a layer Sfc6；Should Network is trained, and is a single example by the incoming frame (image or light stream) of each video, using pre-training net Network；When being classified, each image (or light stream of frame) is initially treated as independent；N for a videofFrame generates NfClassification determines.
Wherein, space-time symbiosis decoding, including by the common united room and time feature occurring, using DCNNs side Method pass through calculate symbiosis combine room and time network convolutional layer, specifically, allow time-space network n-th layer Feature Mapping ForWithdnIt is the number of the dimension of Feature Mapping, calculate Feature Mapping combination
WithRefer to the local feature vectors of the room and time stream on position (i, j), vector quantization is grasped Make,Symbiosis feature as position (i, j)；Therefore, in the Symbiotic Model of each locus, The visual movement of apposition computing capture, maximum pond is applied to all of local coder vector；Pi,jCreate last character representationFinally, L2Standardization is applied to coding vector；
Space-time bilinearity DCNN feature combines fc6 space-time characteristic and is used for double fluid early stage fusion, and this enables us to combine office Portion and the room and time information of the overall situation.
Fusion comprises the steps：
(1) early stage merges：Using two independent sequential networks and spatial network SoWithBinary-flow network be used for action know Not, combine the information of double fluid in early stage, by merging fc6 output, Sfc6And Tfc6, fc6 is first full articulamentum, is often used in Single feature is extracted from DCNNs；We are referred to as this network, and double fluid is early to be melted；
(2) carry out merging in early days, the method then merging double fluid using bilinearity DCNN, room and time information phase is tied Close；The combination initial data pretreatment of the layer by being fully connected, determine classified counting object, carry out point on calculating object Class.
The principle of SVM classifier is：
If linear separability sample set and be (xi,yi), i=1 ..., n, x ∈ Rd, y ∈ {+1, -1 } is category label, then
W x+b=0
It is the classifying face equation of SVM classifier；
In classification, in order that classifying face is correctly classified by all samples and class interval reaches maximum, under needing to meet Two, face condition：
Φ (x)=min (wTw)
yi(w·xi+b)-1≥0
Can be obtained by optimal classification surface by solving this constrained optimization problems, and cross nearest from classifying face in two class samples Put and be just so that, parallel to the training sample on the hyperplane of optimal classification surface, those special sample that in formula, equal sign is set up, Because they support optimal classification surface, therefore it is referred to as support vector；Fusion output is input to SVM as characteristic vector divide Among class device, obtain final classification result.
Fig. 2 is a kind of video data of the monkey class of the video image classifier method based on space-time symbiosis binary-flow network of the present invention Collection.Including image and Optic flow information, data set is made up of 100 kinds of monkey class video sets；This data set is divided into training set and test set. Record monkey class video in certain distance, this data set has larger challenge, such as large-scale camera motion changes and suitable Big attitudes vibration；
Fig. 3 is that a kind of use T-SNE of the video image classifier method based on space-time symbiosis binary-flow network of the present invention is visual Change qualitative evaluation.Visualization technique based on dimensionality reduction is using t- distribution neighborhood embedded mobile GIS (t-SNE) it can be seen that passing through to make Formed with the less detached cluster of the method for symbiosis, with detached cluster often more closely together with, better profit from Temporal information.
Fig. 4 is a kind of space-time symbiosis concept map of the video image classifier method based on space-time symbiosis binary-flow network of the present invention Method.Including by the common united room and time feature occurring, pass through calculating symbiosis combination using DCNNs method empty Between and time network convolutional layer, specifically, allow the Feature Mapping of n-th layer of time-space network beWithdnIt is the number of the dimension of Feature Mapping, calculate Feature Mapping combination
WithRefer to the local feature vectors of the room and time stream on position (i, j), vector quantization is grasped Make,Symbiosis feature as position (i, j)；Therefore, the vision of apposition computing capture and motion exist The Symbiotic Model of each locus, maximum pond is applied to all of local coder vector；Pi,jCreate last character representationFinally, L2Standardization is applied to coding vector；
Space-time bilinearity DCNN feature combines fc6 space-time characteristic and is used for double fluid early stage fusion, and this enables us to combine office Portion and the room and time information of the overall situation.
Fig. 5 is that a kind of early stage convergence strategy of the video image classifier method based on space-time symbiosis binary-flow network of the present invention is shown It is intended to.Fusion comprises the steps：
(1) early stage merges
Using two independent sequential networks and spatial network SoWithBinary-flow network be used for action recognition, early stage tie Close the information of two streams, by merging fc6 output, Sfc6And Tfc6, fc6 is first full articulamentum, is often used in from DCNNs Extract single feature；We are referred to as double-current (early melting) this modification network；
(2) carrying out early stage merges, and the combination initial data pretreatment of the layer by being fully connected, determines classified counting pair As, calculate object on classified；Then the method merging double fluid using bilinearity DCNN, room and time information phase is tied Close.
Fig. 6 is a kind of monkey class example positioning of video image classifier method based on space-time symbiosis binary-flow network of the present invention Figure.In most of the cases, the position of monkey class in image can accurately be navigated to.But work as the stricture of vagina occurring in picture obscuring Reason, when mixed and disorderly object and occlusion condition, the positioning of video image makes a mistake.
For those skilled in the art, the present invention is not restricted to the details of above-described embodiment, in the essence without departing substantially from the present invention In the case of god and scope, the present invention can be realized with other concrete forms.Additionally, those skilled in the art can be to this Bright carry out various change and modification without departing from the spirit and scope of the present invention, these improve and modification also should be regarded as the present invention's Protection domain.Therefore, all changes that claims are intended to be construed to including preferred embodiment and fall into the scope of the invention More and modification.

Claims (10)

1. a kind of video image classifier method based on space-time symbiosis binary-flow network is it is characterised in that mainly include data input (1)；Space-time binary-flow network (two)；Merge (three)；SVM classifier (four).
2. based on a kind of video image classifier method based on space-time symbiosis binary-flow network described in claims 1, its feature It is, including using the early binary-flow network binding time melting and spatial information (space-time symbiosis) method, using the video counts of monkey class According to collection, use more frames (i.e. more spatial datas) to produce a significant precision from each video and improve；Space and when Between information combination, both form complementation, and precision reaches 65.8%.
3. based on a kind of video image classifier method based on space-time symbiosis binary-flow network described in claims 1, its feature It is, the visualization technique based on dimensionality reduction is drawn using t- distribution neighborhood embedded mobile GIS (t-SNE), by using the method for symbiosis Less detached cluster is formed, with detached cluster often more closely together with, better profit from temporal information.
4. based on the data input () described in claims 1 it is characterised in that including image and Optic flow information, data set It is made up of 100 kinds of monkey class video sets；This data set is divided into training set and test set；Record monkey class video in certain distance, should Data set has larger challenge, and such as large-scale camera motion changes and sizable attitudes vibration；
There is provided data below for each class (monkey kind)：There is the video clipping of active annotations, sound clip, bounding box, with And classification and distributing position.
5. based on the test described in claims 4, each video clipping is tested by the way of 5 frames (FPS) per second, calculates every The light stream of 5 frames carrys out computational efficiency.
6. based on the space-time binary-flow network (two) described in claims 1 it is characterised in that including time network, spatial network, And space-time symbiosis decoding.
7. based on the time network described in claims 6 and spatial network it is characterised in that including
(1) sequential networkIt is used as level stream Ox, vertically flow OySize O with light streammagInput combine to form one single Optical signature mapping O ∈ Rh×w×3, wherein h × w is the size of Feature Mapping (image)；
(2) spatial networkRGB frame (image) is used as input；
WithAll using DCNN structure, form S by 5 convolutional layersc1,Sc2,…,Sc5, next to that being fully connected a layer Sfc6；This network It is trained, be a single example by the incoming frame (image or light stream) of each video, using pre-training network；? When being classified, each image (or light stream of frame) is initially treated as independent；N for a videofFrame generates NfClassification Determine.
8. decoded based on the space-time symbiosis described in claims 6 it is characterised in that including by the common united sky occurring Between and temporal characteristics, using DCNNs method pass through calculate symbiosis combine room and time network convolutional layer, specifically, allow The Feature Mapping of the n-th layer of time-space network isWithdnIt is the number of the dimension of Feature Mapping, Calculate Feature Mapping combination
P i , j = v e c ( S i , j n T i , j n T )
WithRefer to the local feature vectors of the room and time stream on position (i, j), carry out vector quantization behaviour Make,Symbiosis feature as position (i, j)；Therefore, in the Symbiotic Model of each locus, The visual movement of apposition computing capture, maximum pond is applied to all of local coder vector；Pi,jCreate last character representationFinally, L2Standardization is applied to coding vector；
Space-time bilinearity DCNN feature combine fc6 space-time characteristic be used for double fluid early stage merge, this enable us to reference to local and The room and time information of the overall situation.
9. based on the fusion (three) described in claims 1 it is characterised in that comprising the steps：
(1) early stage merges
Using two independent sequential networks and spatial network SoWithBinary-flow network be used for action recognition, combine double in early stage The information of stream, by merging fc6 output, Sfc6And Tfc6, fc6 is first full articulamentum, is often used in extracting list from DCNNs Feature；We are referred to as this network, and double fluid is early to be melted；
(2) carry out merging in early days, the method then merging double fluid using bilinearity DCNN, room and time information combines；
The combination initial data pretreatment of the layer by being fully connected, determine classified counting object, carry out point on calculating object Class.
10. based on the SVM classifier (four) described in claims 1 it is characterised in that the principle including SVM classifier is：
If linear separability sample set and be (xi,yi), i=1 ..., n, x ∈ Rd, y ∈ {+1, -1 } is category label, then
W x+b=0
It is the classifying face equation of SVM classifier；
In classification, in order that classifying face is correctly classified by all samples and class interval reaches maximum, need to meet following two Individual condition：
Φ (x)=min (wTw)
yi(w·xi+b)-1≥0
Can be obtained by optimal classification surface by solving this constrained optimization problems, and cross two class samples in from the nearest point of classifying face and Just it is so that those special sample that in formula, equal sign is set up parallel to the training sample on the hyperplane of optimal classification surface, because They support optimal classification surface, are therefore referred to as support vector；It is input to SVM classifier using merging output as characteristic vector Among, obtain final classification result.
CN201610794689.9A 2016-08-31 2016-08-31 A kind of video image classifier method based on space-time symbiosis binary-flow network Pending CN106469314A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201610794689.9A CN106469314A (en) 2016-08-31 2016-08-31 A kind of video image classifier method based on space-time symbiosis binary-flow network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201610794689.9A CN106469314A (en) 2016-08-31 2016-08-31 A kind of video image classifier method based on space-time symbiosis binary-flow network

Publications (1)

Publication Number Publication Date
CN106469314A true CN106469314A (en) 2017-03-01

Family

ID=58230424

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201610794689.9A Pending CN106469314A (en) 2016-08-31 2016-08-31 A kind of video image classifier method based on space-time symbiosis binary-flow network

Country Status (1)

Country Link
CN (1) CN106469314A (en)

Cited By (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107516084A (en) * 2017-08-30 2017-12-26 中国人民解放军国防科技大学 Internet video author identity identification method based on multi-feature fusion
CN107609460A (en) * 2017-05-24 2018-01-19 南京邮电大学 A kind of Human bodys' response method for merging space-time dual-network stream and attention mechanism
CN107945146A (en) * 2017-11-23 2018-04-20 南京信息工程大学 A kind of space-time Satellite Images Fusion method based on depth convolutional neural networks
CN108334901A (en) * 2018-01-30 2018-07-27 福州大学 A kind of flowers image classification method of the convolutional neural networks of combination salient region
CN109190479A (en) * 2018-08-04 2019-01-11 台州学院 A kind of video sequence expression recognition method based on interacting depth study
CN109325430A (en) * 2018-09-11 2019-02-12 北京飞搜科技有限公司 Real-time Activity recognition method and system
CN109889849A (en) * 2019-01-30 2019-06-14 北京市商汤科技开发有限公司 Video generation method, device, medium and equipment
CN110012547A (en) * 2019-04-12 2019-07-12 电子科技大学 A kind of method of user-association in symbiosis network
CN110110648A (en) * 2019-04-30 2019-08-09 北京航空航天大学 Method is nominated in view-based access control model perception and the movement of artificial intelligence
CN110351523A (en) * 2019-07-22 2019-10-18 常州机电职业技术学院 Building video monitoring system and video monitoring adjustment method
CN110516536A (en) * 2019-07-12 2019-11-29 杭州电子科技大学 A kind of Weakly supervised video behavior detection method for activating figure complementary based on timing classification
CN111027377A (en) * 2019-10-30 2020-04-17 杭州电子科技大学 Double-flow neural network time sequence action positioning method
CN111402200A (en) * 2020-02-18 2020-07-10 江苏大学 Fried food detection system based on symbiotic double-current convolution network and digital image
CN111428066A (en) * 2020-04-24 2020-07-17 南京图格医疗科技有限公司 Method for classifying and segmenting lesion image based on convolutional neural network
CN111753782A (en) * 2020-06-30 2020-10-09 西安深信科创信息技术有限公司 False face detection method and device based on double-current network and electronic equipment
CN112364852A (en) * 2021-01-13 2021-02-12 成都考拉悠然科技有限公司 Action video segment extraction method fusing global information
CN112989955A (en) * 2021-02-20 2021-06-18 北方工业大学 Method for recognizing human body actions based on space-time double-current heterogeneous grafting convolutional neural network

Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105550699A (en) * 2015-12-08 2016-05-04 北京工业大学 CNN-based video identification and classification method through time-space significant information fusion

Patent Citations (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105550699A (en) * 2015-12-08 2016-05-04 北京工业大学 CNN-based video identification and classification method through time-space significant information fusion

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
ZONGYUAN GE 等: "Exploiting Temporal Information for DCNN-based Fine-Grained Object Classification", 《HTTPS://ARXIV.ORG/PDF/:1608.00486V1.PDF》 *

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107609460A (en) * 2017-05-24 2018-01-19 南京邮电大学 A kind of Human bodys' response method for merging space-time dual-network stream and attention mechanism
CN107609460B (en) * 2017-05-24 2021-02-02 南京邮电大学 Human body behavior recognition method integrating space-time dual network flow and attention mechanism
CN107516084A (en) * 2017-08-30 2017-12-26 中国人民解放军国防科技大学 Internet video author identity identification method based on multi-feature fusion
CN107516084B (en) * 2017-08-30 2020-01-17 中国人民解放军国防科技大学 Internet video author identity identification method based on multi-feature fusion
CN107945146B (en) * 2017-11-23 2021-08-03 南京信息工程大学 Space-time satellite image fusion method based on deep convolutional neural network
CN107945146A (en) * 2017-11-23 2018-04-20 南京信息工程大学 A kind of space-time Satellite Images Fusion method based on depth convolutional neural networks
CN108334901A (en) * 2018-01-30 2018-07-27 福州大学 A kind of flowers image classification method of the convolutional neural networks of combination salient region
CN109190479A (en) * 2018-08-04 2019-01-11 台州学院 A kind of video sequence expression recognition method based on interacting depth study
CN109325430A (en) * 2018-09-11 2019-02-12 北京飞搜科技有限公司 Real-time Activity recognition method and system
CN109325430B (en) * 2018-09-11 2021-08-20 苏州飞搜科技有限公司 Real-time behavior identification method and system
CN109889849A (en) * 2019-01-30 2019-06-14 北京市商汤科技开发有限公司 Video generation method, device, medium and equipment
CN109889849B (en) * 2019-01-30 2022-02-25 北京市商汤科技开发有限公司 Video generation method, device, medium and equipment
CN110012547B (en) * 2019-04-12 2021-04-02 电子科技大学 User association method in symbiotic network
CN110012547A (en) * 2019-04-12 2019-07-12 电子科技大学 A kind of method of user-association in symbiosis network
CN110110648A (en) * 2019-04-30 2019-08-09 北京航空航天大学 Method is nominated in view-based access control model perception and the movement of artificial intelligence
CN110110648B (en) * 2019-04-30 2020-03-17 北京航空航天大学 Action nomination method based on visual perception and artificial intelligence
CN110516536A (en) * 2019-07-12 2019-11-29 杭州电子科技大学 A kind of Weakly supervised video behavior detection method for activating figure complementary based on timing classification
CN110516536B (en) * 2019-07-12 2022-03-18 杭州电子科技大学 Weak supervision video behavior detection method based on time sequence class activation graph complementation
CN110351523A (en) * 2019-07-22 2019-10-18 常州机电职业技术学院 Building video monitoring system and video monitoring adjustment method
CN111027377A (en) * 2019-10-30 2020-04-17 杭州电子科技大学 Double-flow neural network time sequence action positioning method
CN111027377B (en) * 2019-10-30 2021-06-04 杭州电子科技大学 Double-flow neural network time sequence action positioning method
CN111402200A (en) * 2020-02-18 2020-07-10 江苏大学 Fried food detection system based on symbiotic double-current convolution network and digital image
CN111402200B (en) * 2020-02-18 2021-12-21 江苏大学 Fried food detection system based on symbiotic double-current convolution network and digital image
CN111428066A (en) * 2020-04-24 2020-07-17 南京图格医疗科技有限公司 Method for classifying and segmenting lesion image based on convolutional neural network
CN111753782A (en) * 2020-06-30 2020-10-09 西安深信科创信息技术有限公司 False face detection method and device based on double-current network and electronic equipment
CN111753782B (en) * 2020-06-30 2023-02-10 西安深信科创信息技术有限公司 False face detection method and device based on double-current network and electronic equipment
CN112364852A (en) * 2021-01-13 2021-02-12 成都考拉悠然科技有限公司 Action video segment extraction method fusing global information
CN112989955A (en) * 2021-02-20 2021-06-18 北方工业大学 Method for recognizing human body actions based on space-time double-current heterogeneous grafting convolutional neural network
CN112989955B (en) * 2021-02-20 2023-09-29 北方工业大学 Human body action recognition method based on space-time double-flow heterogeneous grafting convolutional neural network

Similar Documents

Publication Publication Date Title
CN106469314A (en) A kind of video image classifier method based on space-time symbiosis binary-flow network
CN112541501B (en) Scene character recognition method based on visual language modeling network
CN110879959B (en) Method and device for generating data set, and testing method and testing device using same
CN105678216A (en) Spatio-temporal data stream video behavior recognition method based on deep learning
CN110489395A (en) Automatically the method for multi-source heterogeneous data knowledge is obtained
CN107679462A (en) A kind of depth multiple features fusion sorting technique based on small echo
CN110119757A (en) Model training method, video category detection method, device, electronic equipment and computer-readable medium
CN109508375A (en) A kind of social affective classification method based on multi-modal fusion
CN108229338A (en) A kind of video behavior recognition methods based on depth convolution feature
CN105550699A (en) CNN-based video identification and classification method through time-space significant information fusion
CN110263822B (en) Image emotion analysis method based on multi-task learning mode
CN106845329A (en) A kind of action identification method based on depth convolution feature multichannel pyramid pond
CN111898439A (en) Deep learning-based traffic scene joint target detection and semantic segmentation method
CN110852295B (en) Video behavior recognition method based on multitasking supervised learning
CN110781850A (en) Semantic segmentation system and method for road recognition, and computer storage medium
CN110008961A (en) Text real-time identification method, device, computer equipment and storage medium
CN113297370A (en) End-to-end multi-modal question-answering method and system based on multi-interaction attention
CN114443899A (en) Video classification method, device, equipment and medium
WO2021050769A1 (en) Spatio-temporal interactions for video understanding
CN116108215A (en) Cross-modal big data retrieval method and system based on depth fusion
CN113657272B (en) Micro video classification method and system based on missing data completion
CN114677536A (en) Pre-training method and device based on Transformer structure
CN113743389A (en) Facial expression recognition method and device and electronic equipment
CN115471901A (en) Multi-pose face frontization method and system based on generation of confrontation network
Zahedi et al. Robust sign language recognition system using ToF depth cameras

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication

Application publication date: 20170301

RJ01 Rejection of invention patent application after publication