CN109583334A - A kind of action identification method and its system based on space time correlation neural network - Google Patents
A kind of action identification method and its system based on space time correlation neural network Download PDFInfo
- Publication number
- CN109583334A CN109583334A CN201811368191.1A CN201811368191A CN109583334A CN 109583334 A CN109583334 A CN 109583334A CN 201811368191 A CN201811368191 A CN 201811368191A CN 109583334 A CN109583334 A CN 109583334A
- Authority
- CN
- China
- Prior art keywords
- neural network
- module
- space
- network module
- time correlation
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 168
- 238000000034 method Methods 0.000 title claims abstract description 40
- 230000009471 action Effects 0.000 title claims abstract description 25
- 239000000203 mixture Substances 0.000 claims abstract description 10
- 238000012549 training Methods 0.000 claims description 36
- 210000005036 nerve Anatomy 0.000 claims description 23
- 230000002708 enhancing effect Effects 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 9
- 238000010276 construction Methods 0.000 claims description 7
- 230000008569 process Effects 0.000 claims description 6
- 230000000717 retained effect Effects 0.000 claims description 6
- 230000004048 modification Effects 0.000 claims description 4
- 238000012986 modification Methods 0.000 claims description 4
- 210000004218 nerve net Anatomy 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 230000011218 segmentation Effects 0.000 claims description 3
- 238000011478 gradient descent method Methods 0.000 claims 1
- 230000002123 temporal effect Effects 0.000 abstract description 9
- 239000000284 extract Substances 0.000 abstract description 3
- 238000013461 design Methods 0.000 abstract description 2
- 238000000605 extraction Methods 0.000 abstract description 2
- 230000008859 change Effects 0.000 description 3
- 241000208340 Araliaceae Species 0.000 description 2
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 2
- 235000003140 Panax quinquefolius Nutrition 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 235000008434 ginseng Nutrition 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000006855 networking Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Computational Linguistics (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of action identification method based on space time correlation neural network, is trained to obtain space time correlation neural network by the video data for having marked action classification to magnanimity.The network is by the action recognition model of three parts (space neural network module, consecutive frame are associated with neural network module, relating module) composition.The invention can effectively extract the spatial information of video actions and the temporal information of movement, and the extraction of video actions space time information can be better achieved.The space time correlation neural network of design can be realized to be learnt end to end, has good performance in the accuracy rate and speed of action recognition.
Description
Technical field
The present invention relates to artificial intelligence fields, know more particularly, to a kind of movement based on space time correlation neural network
Other method and its system.
Background technique
For the action recognition technology of early stage based on the method for artificial design features, effect is general.Recently as depth
Extensive use of the learning method in computer vision field forms a series of action recognition side based on neural network
Method achieves bigger success.Smionyan in 2014 et al. proposes double-current convolutional neural networks, is replaced using light stream
Temporal information in video achieves bigger success.2016 and Wang etc. devises a kind of time slice network, the net
Network structure can be modeled for a long time by the way that video frame to be segmented, the prolonged action message of acquisition that can be relatively good.
Although binary-flow network achieves certain effect in action recognition, which uses light stream as movement
The characterization of temporal information has Railway Project.The temporal information that can one side light stream represent in movement remains to be discussed, another party
The calculating needs of face light stream take a substantial amount of time, and real-time application scenarios, the light streams such as monitoring are not available, the party is limited
The timeliness of method.Therefore this method can not make a breakthrough in practical application.
Summary of the invention
The present invention be overcome be not able to satisfy described in the above-mentioned prior art all application places and calculating ratio it is relatively time-consuming lack
It falls into, a kind of action identification method and its system based on space time correlation neural network is provided.
The present invention is directed to solve above-mentioned technical problem at least to a certain extent.
In order to solve the above technical problems, technical scheme is as follows: a kind of dynamic based on space time correlation neural network
Make recognition methods, step includes:
S1: construction and training space neural network module；
S2: construction is associated with neural network module with training consecutive frame, uses relating module connection space neural network module
Neural network module is associated with consecutive frame；
S3: training is associated with the space time correlation mind that neural network module is combined by space neural network module with consecutive frame
Through network system；
S4: video input to be measured is subjected to action recognition into trained space time correlation neural network module.
The present invention is closed by construction with the space-time that relating module closes networking module come connection space network module and consecutive frame
Connection neural network module extracts spatial information and temporal information, to accurately carry out the classification of motion to video to be measured.
Preferably, detailed process is as follows by the step S1:
S1.1: being equally divided into 3 sections for each training video, and every section randomly selects a frame, uses the three frames video as space
The input of neural network module, while three frame videos are handled using data enhancing processing technique；
S1.2: building space neural network module, using the good convolutional network model of pre-training to space neural network mould
Block is initialized；
S1.3: the Output Size of the full articulamentum of modification space neural network module protects the class number of itself and movement
It holds consistent；
S1.4: for each video, three video frames of selection are sequentially input, each video frame obtains classification results can
To be expressed as Pre_i, wherein i=1,2,3；
S1.5: the classification results of three frame images are averaged, and as the expression of the final video, can be indicated are as follows:
S1.6: the classification results of video and given label are compared, using the method for stochastic gradient descent to whole
The parameter of a space neural network module is updated, and realizes the training of space neural network module；
S1.7: after the completion of the training of space neural network module, the parameter of retaining space neural network module, for space-time
It is associated with the initialization of nerve network system.
Preferably, detailed process is as follows by the step S2:
S2.1: each training video is equally divided into 3 sections, every section randomly selects continuous 5 frame as space time correlation nerve net
The input of network, while being handled using data enhancing, random cropping and flip horizontal are carried out to each frame image；
S2.2: building consecutive frame association neural network module, and consecutive frame is associated with neural network module structure and space nerve
Network module structure is almost the same, only removes the convolutional layer of space neural network module first layer；
S2.3: building relating module is believed for integrating the information between consecutive frame from association is extracted between consecutive frame
Breath, as the input of consecutive frame association neural network module, relating module is by a three dimensional convolution kernel and a maximum pond
Layer composition；
S2.4: the feature of space neural network module is converted into consecutive frame linked character using relating module；
S2.5: the parameter constant of fixed space neural network module, it is initial using the network parameter retained in step S1.7
Change space neural network module, training consecutive frame is associated with neural network module；
S2.6: the consecutive frame that three are segmented is sequentially inputted in network, and every section of consecutive frame is associated with by consecutive frame
The output result of neural network module can be expressed as Pre_i, wherein i=1, and 2,3；
S2.7: the classification results that three are segmented are averaged, and as the expression of the final video, can be indicated are as follows:
S2.8: the classification results of video and given label are compared, using the method for stochastic gradient descent to whole
The parameter of a consecutive frame association neural network module and relating module is updated, realize consecutive frame association neural network module with
The training of relating module；
S2.9: consecutive frame is associated with after the completion of neural network module training, retains the ginseng of consecutive frame association neural network module
Several and relating module parameter, the initialization for space time correlation neural network.
Preferably, detailed process is as follows by the step S3:
S3.1: building space time correlation neural network, space time correlation neural network is by space neural network module and consecutive frame
Related network is combined, and the output of first convolutional layer of space neural network module is made after the processing of relating module
The input of first residual error module of neural network module is associated with for consecutive frame；
S3.2: space time correlation neural network is initialized using the network parameter retained in S1.7 and S2.9, so
After start to be trained entire space time correlation neural network, realize the optimization of parameter；
S3.3: each video is equally divided into 3 sections, every section randomly selects continuous 5 frame as space time correlation neural network
Input, while being handled using data enhancing, random cropping and flip horizontal carried out to each frame image；
S3.4: the consecutive frame that three are segmented is sequentially inputted in network, and every section of consecutive frame is associated with by consecutive frame
The output result of neural network module is expressed as Pre_ri(i=1,2,3) passes through the output result table of spatial convoluted neural network
It is shown as Pre_si(i=1,2,3)；
S3.5: by output that three are segmented as a result, and the spatial convoluted neural network of each segmentation be associated with consecutive frame
The output result of neural network module blends, and as the classification score of the final video, can indicate are as follows:
S3.6: the classification results of video and given label are compared, using the method for stochastic gradient descent to whole
The parameter of a space time correlation neural network is updated, and realizes the training of space time correlation neural network；
S3.7: trained space time correlation neural network is for the classification to video actions.
In addition a kind of scheme met under same inventive concept has: a kind of movement knowledge based on space time correlation neural network
The system of other method, including space neural network module, consecutive frame association neural network module and relating module；
Space neural network module extracts the spatial information in video；
Consecutive frame association neural network module is used to extract the temporal information in video；
Relating module is used to merge several space neural network modules with several consecutive frames association neural network module
To space time correlation neural network.
Preferably, the space neural network module includes convolutional layer, several residual error modules, pond layer and full connection
Layer；
The consecutive frame association neural network module includes several residual error modules, pond layer and full articulamentum；
The output of first convolutional layer of space neural network module is used as consecutive frame after the processing of relating module
It is associated with the input of first residual error module of neural network module.
Preferably, the space time correlation nerve network system is to be closed several by space neural network module, consecutive frame
Join the network module composition of neural network module and relating module composition, the output of the network module of the composition is by average
Pond obtains classification results.
Compared with prior art, the beneficial effect of technical solution of the present invention is: the invention proposes a kind of new end-to-end
Binary-flow network structure, i.e. space time correlation neural network.Space time correlation neural network is made of Liang Ge branch, is space networks respectively
Network and consecutive frame related network, for extracting spatial information and temporal information in video.Relating module is devised to connect sky
Between network net is associated with consecutive frame, the related information between consecutive frame can be used as movement in temporal information expression.The present invention
It can effectively be proposed from video to spatial information and temporal information, so that capture is for accurately expressing the space-time of video features
Information, for the classification of motion in video, this method has relatively good performance, being capable of accurately and quickly realization movement
Classification.
Detailed description of the invention
Fig. 1 is flow chart of the invention.
Fig. 2 is space time correlation nerve network system of the present invention；
Fig. 3 is relating module of the present invention.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent；
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment 1
Action identification method flow chart based on space time correlation neural network as shown in Figure 1, step include:
S1: construction and training space neural network module；
S1.1: being equally divided into 3 sections for each training video, and every section randomly selects a frame, uses the three frames video as space
The input of neural network module, while being handled using data enhancing, i.e., random cropping and flip horizontal are carried out to each frame image；
S1.2: building space neural network module, using the good convolutional network model of pre-training to space neural network mould
Block is initialized；
S1.3: the Output Size of the full articulamentum of modification space neural network module protects the class number of itself and movement
It holds consistent；
S1.4: for each video, three video frames of selection are sequentially input, each video frame obtains classification results can
To be expressed as Pre_i, wherein i=1,2,3；
S1.5: the classification results of three frame images are averaged, and as the expression of the final video, can be indicated are as follows:
S1.6: the classification results of video and given label are compared, using the method for stochastic gradient descent to whole
The parameter of a space neural network module is updated, and realizes the training of space neural network module；
S1.7: after the completion of the training of space neural network module, the parameter of retaining space neural network module, for space-time
It is associated with the initialization of neural network module.
S2: construction is associated with neural network module with training consecutive frame, uses relating module connection space neural network module
Neural network module is associated with consecutive frame；
S2.1: each training video is equally divided into 3 sections, every section randomly selects continuous 5 frame as space time correlation nerve net
The input of network, while being handled using data enhancing, random cropping and flip horizontal are carried out to each frame image；
S2.2: building consecutive frame association neural network module, and consecutive frame is associated with neural network module structure and space nerve
Network module structure is almost the same, only removes the convolutional layer of space neural network module first layer；
S2.3: building relating module is believed for integrating the information between consecutive frame from association is extracted between consecutive frame
Breath, as the input of consecutive frame association neural network module, relating module is by a three dimensional convolution kernel and a maximum pond
Layer composition, structure are as shown in Figure 3；
S2.4: consecutive frame association neural network module is combined into composition with space neural network module structure, by space
The feature extraction of first convolutional layer of neural network module comes out the input for being used as consecutive frame association neural network module, simultaneously will
The output of first residual error module of space neural network module is added to consecutive frame association nerve by relating module as residual error
The same position of network module, forms space time correlation neural network, and the feature of space neural network module is turned using relating module
Change consecutive frame linked character into；
S2.5: the parameter constant of fixed space neural network module, it is initial using the network parameter retained in step S1.7
Change space neural network module, training consecutive frame is associated with neural network module；
S2.6: the consecutive frame that three are segmented is sequentially inputted in network, and every section of consecutive frame is associated with by consecutive frame
The output result of neural network module can be expressed as Pre_i, wherein i=1, and 2,3；
S2.7: the classification results that three are segmented are averaged, and as the expression of the final video, can be indicated are as follows:
S2.8: the classification results of video and given label are compared, using the method for stochastic gradient descent to whole
The parameter of a consecutive frame association neural network module and relating module is updated, realize consecutive frame association neural network module with
The training of relating module；
S2.9: consecutive frame is associated with after the completion of neural network module training, retains the ginseng of consecutive frame association neural network module
Several and relating module parameter, the initialization for space time correlation neural network.
S3: training is associated with the space time correlation mind that neural network module is combined by space neural network module with consecutive frame
Through network system；
S3.1: building space time correlation nerve network system, space time correlation neural network is by space neural network module and phase
Adjacent frame related network is combined, first convolutional layer of space neural network module output by relating module processing it
The input as first residual error module of consecutive frame association neural network module, specific system diagram are as shown in Figure 2 afterwards；
S3.2: space time correlation neural network is initialized using the network parameter retained in S1.7 and S2.9, so
After start to be trained entire space time correlation neural network, realize the optimization of parameter；
S3.3: each video is equally divided into 3 sections, every section randomly selects continuous 5 frame as space time correlation neural network
Input, while being handled using data enhancing, random cropping and flip horizontal carried out to each frame image；
S3.4: the consecutive frame that three are segmented is sequentially inputted in network, and every section of consecutive frame is associated with by consecutive frame
The output result of neural network module is expressed as Pre_ri(i=1,2,3) passes through the output result table of spatial convoluted neural network
It is shown as Pre_si(i=1,2,3)；
S3.5: by output that three are segmented as a result, and the spatial convoluted neural network of each segmentation be associated with consecutive frame
The output result of neural network module blends, and as the classification score of the final video, can indicate are as follows:
S3.6: the classification results of video and given label are compared, using the method for stochastic gradient descent to whole
The parameter of a space time correlation neural network is updated, and realizes the training of space time correlation neural network；
S3.7: trained space time correlation neural network is for the classification to video actions.
S4: video input to be measured is subjected to action recognition into trained space time correlation nerve network system.
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (7)
1. a kind of action identification method based on space time correlation neural network, which comprises the following steps:
S1: construction and training space neural network module；
S2: construction is associated with neural network module with training consecutive frame, uses relating module connection space neural network module and phase
Adjacent frame is associated with neural network module；
S3: training is associated with the space time correlation nerve net that neural network module is combined by space neural network module with consecutive frame
Network system；
S4: video input to be measured is subjected to action recognition into trained space time correlation nerve network system.
2. the action identification method according to claim 1 based on space time correlation neural network, which is characterized in that the step
Detailed process is as follows by rapid S1:
S1.1: being equally divided into 3 sections for each training video, and every section randomly selects a frame, uses the three frames video as space nerve
The input of network module, while three frame videos are handled using data enhancing processing technique；
S1.2: building space neural network module, using the good convolutional network model of pre-training to space neural network module into
Row initialization；
S1.3: the Output Size of the full articulamentum of modification space neural network module makes the class number of itself and movement keep one
It causes；
S1.4: for each video, sequentially inputting three video frames of selection, and the classification results that each video frame obtains can be with
It is expressed as Pre_i, wherein i=1,2,3；
S1.5: the classification results of three frame images are averaged, and as the expression of the final video, are indicated are as follows:
S1.6: the classification results of video and given label are compared, using the method for stochastic gradient descent to entire sky
Between the parameter of neural network module be updated, realize the training of space neural network module；
S1.7: after the completion of the training of space neural network module, the parameter of retaining space neural network module, for space time correlation
The initialization of nerve network system.
3. the action identification method according to claim 2 based on space time correlation neural network, which is characterized in that the step
Detailed process is as follows by rapid S2:
S2.1: each training video is equally divided into 3 sections, every section randomly selects continuous 5 frame as consecutive frame and is associated with neural network
The input of module, while 5 frame videos are handled using data enhancing processing technique；
S2.2: consecutive frame association neural network module is built；
S2.3: building relating module, from related information is extracted between consecutive frame, is made for integrating the information between consecutive frame
The input of neural network module is associated with for consecutive frame, relating module is by a three dimensional convolution kernel and a maximum pond layer group
At；
S2.4: the feature of space neural network module is converted into consecutive frame linked character using relating module；
S2.5: the parameter constant of fixed space neural network module is initialized empty using the network parameter retained in step S1.7
Between neural network module, training consecutive frame be associated with neural network module；
S2.6: the consecutive frame that three are segmented is sequentially inputted in consecutive frame association neural network module, every section of consecutive frame is logical
The output result for crossing consecutive frame association neural network module can be expressed as Pre_i, wherein i=1, and 2,3；
S2.7: the classification results that three are segmented are averaged, and as the expression of the final video, are indicated are as follows:
S2.8: the classification results of video and given label are compared, using the method for stochastic gradient descent to entire phase
The parameter of adjacent frame association neural network module and relating module is updated, realize consecutive frame association neural network module be associated with
The training of module；
S2.9: consecutive frame be associated with neural network module training after the completion of, retain consecutive frame association neural network module parameter with
And the parameter of relating module, the initialization for space time correlation nerve network system.
4. the action identification method according to claim 3 based on space time correlation neural network, which is characterized in that the step
Detailed process is as follows by rapid S3:
S3.1: building space time correlation nerve network system, space time correlation nerve network system is by space neural network module and phase
Adjacent frame related network is combined, first convolutional layer of space neural network module output by relating module processing it
Input as first residual error module of consecutive frame association neural network module afterwards；
S3.2: space time correlation nerve network system is initialized using the network parameter retained in S1.7 and S2.9, so
After start to be trained entire space time correlation nerve network system, realize the optimization of parameter；
S3.3: each video is equally divided into 3 sections, every section randomly selects continuous 5 frame as space time correlation nerve network system
Input, while using data enhancing processing technique 5 frame videos are handled；
S3.4: the consecutive frame that three are segmented is sequentially inputted in space time correlation nerve network system, every section of consecutive frame passes through
The output result of consecutive frame association neural network module is expressed as Pre_ri(i=1,2,3), passes through space neural network module
Output result is expressed as Pre_si(i=1,2,3)；
S3.5: by output that three are segmented as a result, and the space neural network module of each segmentation nerve is associated with consecutive frame
The output result of network module blends, and as the classification score of the final video, indicates are as follows:
S3.6: the classification results of video and given label are compared, using stochastic gradient descent method to it is entire when
The parameter of null Context nerve network system is updated, and realizes the training of space time correlation nerve network system.
5. a kind of system of the method for any one according to claim 1~4, which is characterized in that including space neural network module, phase
Adjacent frame association neural network module and relating module.
6. a kind of system of action identification method based on space time correlation neural network according to claim 5, feature
It is:
The space neural network module includes convolutional layer, several residual error modules, pond layer and full articulamentum；
The consecutive frame association neural network module includes several residual error modules, pond layer and full articulamentum；
The output of first convolutional layer of space neural network module is used as consecutive frame to be associated with after the processing of relating module
The input of first residual error module of neural network module.
7. a kind of system of action identification method based on space time correlation neural network according to claim 5, feature
Be: the space time correlation nerve network system includes several space neural network modules, consecutive frame association neural network mould
The output of block and relating module, the space time correlation nerve network system obtains classification results by average pond.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811368191.1A CN109583334B (en) | 2018-11-16 | 2018-11-16 | Action recognition method and system based on space-time correlation neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811368191.1A CN109583334B (en) | 2018-11-16 | 2018-11-16 | Action recognition method and system based on space-time correlation neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109583334A true CN109583334A (en) | 2019-04-05 |
| CN109583334B CN109583334B (en) | 2022-11-04 |
Family
ID=65922885
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811368191.1A Active CN109583334B (en) | 2018-11-16 | 2018-11-16 | Action recognition method and system based on space-time correlation neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109583334B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110147876A (en) * | 2019-05-09 | 2019-08-20 | 中山大学 | The neural network and its movement motion generation method of view-based access control model characteristic similarity |
| CN111317462A (en) * | 2020-03-20 | 2020-06-23 | 佛山科学技术学院 | Blood flow imaging method and device based on U-net neural network |
| CN111783506A (en) * | 2019-05-17 | 2020-10-16 | 北京京东尚科信息技术有限公司 | Method and device for determining target characteristics and computer-readable storage medium |
| CN114037930A (en) * | 2021-10-18 | 2022-02-11 | 苏州大学 | Video action recognition method based on space-time enhanced network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862376A (en) * | 2017-10-30 | 2018-03-30 | 中山大学 | A kind of human body image action identification method based on double-current neutral net |
| CN108256562A (en) * | 2018-01-09 | 2018-07-06 | 深圳大学 | Well-marked target detection method and system based on Weakly supervised space-time cascade neural network |
| CN108764128A (en) * | 2018-05-25 | 2018-11-06 | 华中科技大学 | A kind of video actions recognition methods based on sparse time slice network |
| CN109460707A (en) * | 2018-10-08 | 2019-03-12 | 华南理工大学 | A kind of multi-modal action identification method based on deep neural network |
-
2018
- 2018-11-16 CN CN201811368191.1A patent/CN109583334B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107862376A (en) * | 2017-10-30 | 2018-03-30 | 中山大学 | A kind of human body image action identification method based on double-current neutral net |
| CN108256562A (en) * | 2018-01-09 | 2018-07-06 | 深圳大学 | Well-marked target detection method and system based on Weakly supervised space-time cascade neural network |
| CN108764128A (en) * | 2018-05-25 | 2018-11-06 | 华中科技大学 | A kind of video actions recognition methods based on sparse time slice network |
| CN109460707A (en) * | 2018-10-08 | 2019-03-12 | 华南理工大学 | A kind of multi-modal action identification method based on deep neural network |
Non-Patent Citations (1)
| Title |
|---|
| 张怡佳等: "基于双流卷积神经网络的改进人体行为识别算法", 《计算机测量与控制》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110147876A (en) * | 2019-05-09 | 2019-08-20 | 中山大学 | The neural network and its movement motion generation method of view-based access control model characteristic similarity |
| CN110147876B (en) * | 2019-05-09 | 2023-01-03 | 中山大学 | Neural network based on visual feature similarity and action proposal generation method thereof |
| CN111783506A (en) * | 2019-05-17 | 2020-10-16 | 北京京东尚科信息技术有限公司 | Method and device for determining target characteristics and computer-readable storage medium |
| WO2020233427A1 (en) * | 2019-05-17 | 2020-11-26 | 北京京东尚科信息技术有限公司 | Method and apparatus for determining features of target |
| CN111317462A (en) * | 2020-03-20 | 2020-06-23 | 佛山科学技术学院 | Blood flow imaging method and device based on U-net neural network |
| CN111317462B (en) * | 2020-03-20 | 2023-11-03 | 佛山科学技术学院 | Blood flow imaging method and device based on U-net neural network |
| CN114037930A (en) * | 2021-10-18 | 2022-02-11 | 苏州大学 | Video action recognition method based on space-time enhanced network |
| CN114037930B (en) * | 2021-10-18 | 2022-07-12 | 苏州大学 | Video action recognition method based on space-time enhanced network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109583334B (en) | 2022-11-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108830252A (en) | A kind of convolutional neural networks human motion recognition method of amalgamation of global space-time characteristic | |
| CN109583334A (en) | A kind of action identification method and its system based on space time correlation neural network | |
| Zhang et al. | Real-time action recognition with enhanced motion vector CNNs | |
| CN106096568B (en) | A kind of pedestrian's recognition methods again based on CNN and convolution LSTM network | |
| Tran et al. | Two-stream flow-guided convolutional attention networks for action recognition | |
| CN107480178B (en) | Pedestrian re-identification method based on cross-modal comparison of image and video | |
| CN109886358A (en) | Human bodys' response method based on multi-space information fusion convolutional neural networks | |
| CN107862376A (en) | A kind of human body image action identification method based on double-current neutral net | |
| CN110363131A (en) | Anomaly detection method, system and medium based on human skeleton | |
| Xu et al. | Fully-coupled two-stream spatiotemporal networks for extremely low resolution action recognition | |
| CN113239801B (en) | Cross-domain action recognition method based on multi-scale feature learning and multi-level domain alignment | |
| CN105678216A (en) | Spatio-temporal data stream video behavior recognition method based on deep learning | |
| CN110110686A (en) | Based on the human motion recognition methods for losing double-current convolutional neural networks more | |
| CN112288627A (en) | Recognition-oriented low-resolution face image super-resolution method | |
| CN109858407A (en) | A kind of video behavior recognition methods based on much information stream feature and asynchronous fusion | |
| CN109948721A (en) | A kind of video scene classification method based on video presentation | |
| CN110070027A (en) | Pedestrian based on intelligent internet of things system recognition methods again | |
| CN109086707A (en) | A kind of expression method for tracing based on DCNNs-LSTM model | |
| CN107330381A (en) | A kind of face identification method | |
| CN113705384A (en) | Facial expression recognition method considering local space-time characteristics and global time sequence clues | |
| CN115410119A (en) | Violent movement detection method and system based on adaptive generation of training samples | |
| CN109002808B (en) | Human behavior recognition method and system | |
| CN111898458B (en) | Violent video identification method for bimodal task learning based on attention mechanism | |
| Zhai et al. | 3D dual-stream convolutional neural networks with simple recurrent unit network: A new framework for action recognition | |
| CN112487926A (en) | Scenic spot feeding behavior identification method based on space-time diagram convolutional network |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |