CN105740815B - A kind of Human bodys' response method based on depth recurrence stratified condition random field - Google Patents
A kind of Human bodys' response method based on depth recurrence stratified condition random field Download PDFInfo
- Publication number
- CN105740815B CN105740815B CN201610064349.0A CN201610064349A CN105740815B CN 105740815 B CN105740815 B CN 105740815B CN 201610064349 A CN201610064349 A CN 201610064349A CN 105740815 B CN105740815 B CN 105740815B
- Authority
- CN
- China
- Prior art keywords
- behavior
- human
- video segment
- random field
- video
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 26
- 230000004044 response Effects 0.000 title abstract 2
- 230000006399 behavior Effects 0.000 claims abstract description 90
- 238000012549 training Methods 0.000 claims description 24
- 230000002452 interceptive effect Effects 0.000 claims description 23
- 230000006870 function Effects 0.000 claims description 18
- 230000036544 posture Effects 0.000 claims description 14
- 230000009471 action Effects 0.000 claims description 9
- 238000004422 calculation algorithm Methods 0.000 claims description 8
- 238000005381 potential energy Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 230000001808 coupling effect Effects 0.000 claims description 4
- 238000013507 mapping Methods 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 238000005457 optimization Methods 0.000 abstract description 2
- 238000012706 support-vector machine Methods 0.000 abstract description 2
- 238000013145 classification model Methods 0.000 abstract 1
- 230000000694 effects Effects 0.000 description 3
- 238000009795 derivation Methods 0.000 description 2
- 238000012545 processing Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 238000010224 classification analysis Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000000306 recurrent 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of Human bodys' response methods based on depth recurrence stratified condition random field, first, the human body attitude of behavior act main body and the object information that may be interacted with it in the RGB-D video by RGB-D video camera shooting behavior act scene are extracted respectively, using both information as the intermediate layer state of depth recurrence stratified condition random field, the correlation of current state and current all prediction output state set occurred, constructs depth recurrence stratified condition random field models in modeling and forecasting output target-like state layer；Secondly, learning the identification and classification model about human body behavior sequence using the structuring support vector machine classifier of BCFW optimization method driving；Finally, the model parameter obtained according to study and the classification for predicting human body behavior sequence to be tested up to discrimination model.The present invention has significant robustness to behavior act, improves the recognition accuracy of human body behavior act to a certain extent.
Description
Technical Field
The invention relates to a human behavior recognition method, in particular to a human behavior recognition method based on Deep Recursive and Hierarchical Conditional Random Fields (DR-HCRFs), and belongs to the technical field of computer vision behavior recognition.
Background
Human behavior recognition has a very important position in computer vision, and has wide application in the fields of intelligent monitoring, man-machine interaction, sports video processing and the like.
In recent years, a probabilistic graphical model method is mainly used for the behavior recognition research of indoor scenes to perform classification analysis on human behaviors. Common probabilistic graphical models are mainly divided into two structures: a generation model and a decision model. Common generative models are: hidden Markov Models (Hidden Markov Models), Bayesian networks (DBNs), Semi-Markov Models (Semi-Markov Models). The generation of the model requires modeling of the distribution and correlation of the prior information, and when there is a complex correlation between the input variables, the modeling of the joint distribution becomes complex or even inaccurate. Instead, using a decision model to model conditional probabilities, accurate and efficient inferences can be derived. For example: hidden conditional random fields (Hidden CRFs), cyclic conditional random fields (Loopy CRFs).
Furthermore, existing behavior recognition research focuses on adding semantic context information, such as object and behavior, behavior and behavior context information, to the existing technology. Experiments prove that the semantic information can be used as a hidden state in a discriminant model to improve the accuracy of behavior recognition. "recurrent belief estimation over CRFs in RGB-D activity videos" published on RSS in 2015 by O.Sener and A.Saxena.rCRF, human posture features and object inspiration information are extracted from RGB videos, and the confidence relation between human behaviors is calculated by adding a conditional random field model of recursive Bayesian estimation. Modeling 3D environments through hidden human context published by PAMI in 2015 in y.jiang, h.s.koppla and a.saxena propose a conditional random field model in an infinitely hidden state, which has a significant effect on the processing of a large number of human postures and interactive objects in a 3D environment. Chatzis and Y.Demiris published in 2013 on PAMI, The "The infinite-order conditional random field model for sequential data modeling", proposes an infinite order conditional random field to model serialized data, and simultaneously uses a sequence memory (sequence memorizer) method to model infinite order correlation in tag sequences.
The existing behavior recognition method based on the probability map model does not consider the high-order correlation between the internal representation of the target state and the state at the same time, and still has the problem of low recognition accuracy.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: a human body behavior recognition method based on a deep recursive hierarchical conditional random field is provided, a human body posture and an interactive object are used as intermediate representation states of a predicted target state, and a deep recursive hierarchical conditional random field model containing input data, the intermediate states and the target predicted state is constructed.
The invention adopts the following technical scheme for solving the technical problems:
a human behavior recognition method based on a deep recursive hierarchical conditional random field comprises the following steps:
step 1, acquiring an RGB-D training video sample of human behaviors, wherein the RGB-D training video sample comprises RGB video information, depth information and human skeleton information, combining the RGB video information and the human skeleton information, extracting human posture characteristics, shape and position characteristics of an interactive object and relative position characteristics of a human body and the interactive object from the RGB video information and the human skeleton information, and connecting the characteristics in series to obtain behavior representation characteristics;
step 2, according to the behavior representation characteristics obtained in the step 1, constructing a fully connected probabilistic graph model formed by linking three parts of behavior representation characteristics, intermediate states formed by human postures and interactive objects and behavior prediction labels in the current video segment, and establishing a deep recursive hierarchical conditional random field model of the current video segment by combining the behavior prediction labels from the first video segment to the previous video segment of the current video segment in the training video sample;
step 3, converting the deep recursive hierarchical conditional random field model established in the step 2 into a first-order linear chain element random field model by using an average field approximation algorithm;
step 4, learning the parameters of the first-order linear chain element random field model obtained in the step 3 by using a maximum-interval algorithm;
and 5, identifying a behavior prediction label corresponding to the test video sample according to the first-order linear chain element random field model obtained in the step 3 and the parameters obtained by learning in the step 4.
Preferably, the potential energy function Ψ (y, h, o, x; ω) of the deep recursive hierarchical conditional random field model is:
where T is 1, …, T represents the T-th video segment of the training video sample, ω1、ω2、ω3、ω4All represent parameters of the model, ht、ot、ytRespectively representing the human body posture, the interactive object and the behavior prediction tag of the t-th video segment;denotes xtAnd ht、otDependence of (c), phi (x)t) Representing a behavior representation feature x in the tth video segmenttA mapping function to a feature space;represents htAnd otThe correlation between the two or more of the three,indicating whether an interactive object s appears in the tth video segment during the action,representing all crossings in the tth video segmentA set of mutual objects, S represents a set of all mutual objects in the training video sample; omega3(yt,ht,ot) Denotes ytAnd ht、otThe coupling property of (a);representing a set of historiesAnd ytThe correlation of (c).
Preferably, the specific process of step 3 is as follows: finding out the optimal behavior prediction label of the current video segment, wherein the optimal behavior prediction labelCan be expressed as:
wherein,the optimal behavior prediction labels of the 1 st video segment to the t-2 th video segment of the training video sample are represented, V and u respectively represent candidate behavior prediction labels in a candidate behavior prediction label set Y ═ { 1.. V }, and V represents the total number of the candidate behavior prediction labels forming the set Y.
Preferably, the calculation expression of the parameter is:where λ represents the equilibrium weight value, ω represents a parameter of the model,represents the optimal behavior prediction label of the ith training video sample, N represents the total number of the training video samples,optimal behavior prediction label representing ith training video sampleAnd an actual behavior tag yiLoss function of difference.
Preferably, the tool for acquiring the RGB-D training video samples of human body behaviors in step 1 is a Kinect depth sensor.
Compared with the prior art, the invention adopting the technical scheme has the following technical effects:
the human body behavior recognition method based on the deep recursion hierarchical conditional random field introduces and increases the high-order correlation between the intermediate structure inside the current behavior action correlation factor and the past behavior action during modeling, has obvious robustness on human body appearance difference, complex scenes, interactive objects and the like contained in the behavior action process, and can improve the recognition accuracy of the human body behavior action to a certain extent.
Drawings
FIG. 1 is a schematic diagram of a deep recursive hierarchical conditional random field proposed by the present invention.
FIG. 2 is a flow chart of the human behavior recognition method based on the deep recursive hierarchical conditional random field according to the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are illustrative only for the purpose of explaining the present invention, and are not to be construed as limiting the present invention.
The specific flow of the human behavior recognition method based on the deep recursive hierarchical conditional random field is shown in fig. 2, and specifically comprises the following steps:
step 1, capturing an RGB-D video sequence of human behaviors by using a Kinect depth sensor, obtaining depth information of a shot scene through obtaining to extract human skeleton structure information of a behavior action subject person, combining two data sources of a human skeleton and the RGB video sequence, extracting human posture characteristics, shape and position characteristics of an interactive object and relative position information of the human body and the object, connecting the information in series for representation, combining to form finally observed behavior representation characteristics, and using the finally observed behavior representation characteristics as subsequent input.
Step 2, constructing input observation data x according to the behavior representation characteristics of the inputtTo human posture htAnd an interactive object otIntermediate states of composition, and final behavior prediction tag ytAnd establishing a corresponding deep recursive hierarchical conditional random field model by using a fully connected probability graph model formed by linking the three parts.
For input observation data x, the probability that its corresponding behavior label is y can be expressed as the following objective function:
the potential energy function of the Ψ (y, h, o, x; ω) depth recursive hierarchical conditional random field model measures the correlation among input observation data x, the human body posture h and the interactive object o in the middle layer, and the behavior prediction label y, and ω is a model parameter. The potential energy function Ψ (y, h, o, x; ω) contains four components, respectively:
A. inputting observation data xtAnd an intermediate state htAnd otDependence of (a):
Ψ1(ht，ot，xt；ω1)＝ω1(ht，ot)·φ(xt) (2)
wherein phi (x)t) Is a mapping function of the input data to a feature space.
B. Human body posture htAnd an interactive object otCorrelation between the two:
wherein,whether an object s appears in the action process in the t video segment or not is shown whenTime indicates occurrence; when in useIndicating that it is not present. S represents a set of all object tags.
C. Behavior prediction label ytAnd an intermediate state htAnd otThe coupling property of (2):
Ψ3(yt,ht,ot；ω3)＝ω3(yt,ht,ot) (4)
D. historical collection of target statesAnd a pre-target state behavior prediction label ytThe correlation of (a):
for currently observed input observation data x (assuming T represents the total length of the input data sequence or the total video segment of the training video sample), in combination with the above equations (2) - (5), a potential energy function of the deep recursive hierarchical conditional random field model can be obtained, which is specifically expressed as follows:
and 3, performing model derivation on the obtained deep recursive hierarchical conditional random field by adopting a mean-field-like algorithm (mean-field-like) so as to reduce the computational complexity of the constructed model and convert the model into a first-order linear chain element random field model.
The purpose of the model derivation is to find the y:
equation (6) is transformed into a backward recursive dynamic programming problem:
where V denotes a predicted target state, and is one element in the target set Y {1, … V }, and V denotes the total number of elements constituting the set Y.
The initial state is as follows:
ζ1(v)＝ω3(y1＝v，h1，o1) (10)
from ζ1(v) Can be regarded as predicting the current input number under the condition that the history is known to predict the optimal stateAccording to the corresponding behavior label. Taking the initial state as the historical state, zeta corresponding to the second time period in the video sequence can be obtainedt(v)：
According to the first and second time periods, zeta corresponding to the third time period can be further obtainedt(v)：
Here, the present invention assumes ytIs not subject to recursive expression ζ3(v) Y in (1)2And y3Influence when y1Obtaining an optimal predicted state valueThen, the above formula can be further expressed as:
is Zeta3(v) The average field of (a) approximates the result. Similarly, ζ at known fourth time periodt(v) Comprises the following steps:
from the point of view of the mean field approximation algorithm, neglecting y1And y2The influence of the fluctuation of (c):
by this approximate optimization method, the present invention can be appliedThe dynamic programming problem is converted into a depth recursive function zetat(v) Expressed simple problems with the worst upper bound on computational complexity. Zetat(v) The infinite order correlation contained in (a) can be expressed as:
and 4, learning the parameters of the constructed random field model by adopting a maximum-interval algorithm (Max-margin).
In the training data setIncludes N training video samples xiAnd a behavior prediction tag sequence yiSamples in a one-to-one correspondence. However, the intermediate states h and o are unknown, and the purpose of model learning in the invention is to find out the optimal model parameter omega, so that the difference between the predicted behavior note and the actual note is the minimum. In order to prevent the overfitting phenomenon, the invention provides a corresponding regular term expression:
wherein λ is a balanced weight value of the data,represents the optimal behavior prediction tag sequence obtained by equation (7).Is a loss function representing the difference between the predicted output behavior sequence and the actual behavior sequence. The concrete expression is as follows:
wherein Z is over a given sequence length ZiNext, z-th value in the sequence. The direct solution of the formula (18) is an N-P problem, and the method carries out marginalized substitution on the original loss function to obtain the upper boundary of the given loss function. The original problem is therefore rewritten as the minimum problem with a conditional constraint to solve the objective function:
wherein the difference valueRelaxation variable ξiThe substitution loss function for the ith data point is shown. Here, the original problem is transformed into a convex problem that can be solved using a structured-Support-Vector Machine (SSVM). But due to YiThe various combination properties inside cause the convex problem to have an exponential condition constraint, and the time spent in the learning process is large. The invention replaces sigma with N piecewise linear pairsi|YiAnd | linear constraint, defining a structured chain loss function corresponding to formula (19):
finally, an equivalent unconstrained formula of formula (17) is obtained:
and finally solving the problem result by a block-coordinate original-dual Frank-Walff (block-coordinate primary-dual Fralnk-Wolfe, BCFW) algorithm.
And 5, predicting behavior labels corresponding to the video sequences for the test data set according to the depth recursive hierarchical conditional random field model and the model parameters obtained by learning.
The above embodiments are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modifications made on the basis of the technical scheme according to the technical idea of the present invention fall within the protection scope of the present invention.
Claims (5)
1. A human behavior recognition method based on a deep recursive hierarchical conditional random field is characterized by comprising the following steps:
step 1, acquiring an RGB-D training video sample of human behaviors, wherein the RGB-D training video sample comprises RGB video information, depth information and human skeleton information, combining the RGB video information and the human skeleton information, extracting human posture characteristics, shape and position characteristics of an interactive object and relative position characteristics of a human body and the interactive object from the RGB video information and the human skeleton information, and connecting the characteristics in series to obtain behavior representation characteristics;
step 2, according to the behavior representation characteristics obtained in the step 1, constructing a fully connected probabilistic graph model formed by linking three parts of behavior representation characteristics, intermediate states formed by human postures and interactive objects and behavior prediction labels in the current video segment, and establishing a deep recursive hierarchical conditional random field model of the current video segment by combining the behavior prediction labels from the first video segment to the previous video segment of the current video segment in the training video sample;
step 3, converting the deep recursive hierarchical conditional random field model established in the step 2 into a first-order linear chain element random field model by using an average field approximation algorithm;
step 4, learning the parameters of the first-order linear chain element random field model obtained in the step 3 by using a maximum-interval algorithm;
and 5, identifying a behavior prediction label corresponding to the test video sample according to the first-order linear chain element random field model obtained in the step 3 and the parameters obtained by learning in the step 4.
2. The method for human behavior recognition based on deep recursive hierarchical conditional random fields according to claim 1, wherein the potential energy function Ψ (y, h, o, x; ω) of the deep recursive hierarchical conditional random field model is:
where T is 1, …, T represents the T-th video segment of the training video sample, ω1、ω2、ω3、ω4All represent parameters of the model, ht、ot、ytRespectively representing the human body posture, the interactive object and the behavior prediction tag of the t-th video segment;denotes xtAnd ht、otDependence of (c), phi (x)t) Representing a behavior representation feature x in the tth video segmenttA mapping function to a feature space;represents htAnd otThe correlation between the two or more of the three,indicating whether an interactive object s appears in the tth video segment during the action,representing a set of all interactive objects in the t video segment, and S represents a set of all interactive objects in the training video sample; omega3(yt,ht,ot) Denotes ytAnd ht、otThe coupling property of (a);representing a set of historiesAnd ytThe correlation of (c).
3. The method for recognizing human body behaviors based on deep recursive hierarchical conditional random fields according to claim 1, wherein the specific process of the step 3 is as follows: finding out the optimal behavior prediction label of the current video segment, wherein the optimal behavior prediction labelCan be expressed as:
wherein, representing the optimal behavior prediction labels of the 1 st video segment to the t-2 th video segment of the training video sample, V and u both representing the candidate behavior prediction labels in a set of candidate behavior prediction labels Y ═ { 1.. V }, V representing the total number of candidate behavior prediction labels constituting the set Y, ω1、ω2、ω3、ω4All represent parameters of the model, ht、ot、ytRespectively representing the human body posture, the interactive object and the behavior prediction label of the t-th video segment,denotes xtAnd ht、otThe dependence of (a) on (b),represents htAnd otThe correlation between the two or more of the three,indicating whether an interactive object s appears in the tth video segment during the action,representing a set of all interactive objects in the t video segment, and S represents a set of all interactive objects in the training video sample; omega3(yt＝v,ht,ot) Denotes ytAnd ht、otThe coupling property of (a);to representAnd yt、yt-1The correlation of (c).
4. The method for human behavior recognition based on deep recursive hierarchical conditional random fields according to claim 1, wherein the computational expression of the parameters is as follows:
where λ represents the equilibrium weight value, ω represents a parameter of the model,represents the optimal behavior prediction label of the ith training video sample, N represents the total number of the training video samples,optimal behavior prediction label representing ith training video sampleAnd an actual behavior tag yiLoss function of difference.
5. The method for human behavior recognition based on deep recursive hierarchical conditional random fields as claimed in claim 1, wherein the tool for obtaining RGB-D training video samples of human behavior in step 1 is a Kinect depth sensor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610064349.0A CN105740815B (en) | 2016-01-29 | 2016-01-29 | A kind of Human bodys' response method based on depth recurrence stratified condition random field |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610064349.0A CN105740815B (en) | 2016-01-29 | 2016-01-29 | A kind of Human bodys' response method based on depth recurrence stratified condition random field |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105740815A CN105740815A (en) | 2016-07-06 |
| CN105740815B true CN105740815B (en) | 2018-12-18 |
Family
ID=56247141
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610064349.0A Active CN105740815B (en) | 2016-01-29 | 2016-01-29 | A kind of Human bodys' response method based on depth recurrence stratified condition random field |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105740815B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107341471B (en) * | 2017-07-04 | 2019-10-01 | 南京邮电大学 | A kind of Human bodys' response method based on Bilayer condition random field |
| CN107491735B (en) * | 2017-07-20 | 2020-08-18 | 浙江工业大学 | Tag and interaction relation joint learning method for human behavior recognition |
| CN111914807B (en) * | 2020-08-18 | 2022-06-28 | 太原理工大学 | Miner behavior identification method based on sensor and skeleton information |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103810496A (en) * | 2014-01-09 | 2014-05-21 | 江南大学 | 3D (three-dimensional) Gaussian space human behavior identifying method based on image depth information |
| CN104598890A (en) * | 2015-01-30 | 2015-05-06 | 南京邮电大学 | Human body behavior recognizing method based on RGB-D video |
| CN104809187A (en) * | 2015-04-20 | 2015-07-29 | 南京邮电大学 | Indoor scene semantic annotation method based on RGB-D data |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7526123B2 (en) * | 2004-02-12 | 2009-04-28 | Nec Laboratories America, Inc. | Estimating facial pose from a sparse representation |
-
2016
- 2016-01-29 CN CN201610064349.0A patent/CN105740815B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103810496A (en) * | 2014-01-09 | 2014-05-21 | 江南大学 | 3D (three-dimensional) Gaussian space human behavior identifying method based on image depth information |
| CN104598890A (en) * | 2015-01-30 | 2015-05-06 | 南京邮电大学 | Human body behavior recognizing method based on RGB-D video |
| CN104809187A (en) * | 2015-04-20 | 2015-07-29 | 南京邮电大学 | Indoor scene semantic annotation method based on RGB-D data |
Non-Patent Citations (1)
| Title |
|---|
| 人体行为识别的条件随机场方法;王媛媛 等;《重庆理工大学学报(自然科学)》;20130630;第27卷(第6期);第93-99、105页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105740815A (en) | 2016-07-06 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liao et al. | Dynamic sign language recognition based on video sequence with BLSTM-3D residual networks | |
| CN108875807B (en) | Image description method based on multiple attention and multiple scales | |
| CN109891897B (en) | Method for analyzing media content | |
| Xu et al. | X-invariant contrastive augmentation and representation learning for semi-supervised skeleton-based action recognition | |
| Zhao et al. | A spatial-temporal attention model for human trajectory prediction. | |
| CN111079646A (en) | Method and system for positioning weak surveillance video time sequence action based on deep learning | |
| CN111310672A (en) | Video emotion recognition method, device and medium based on time sequence multi-model fusion modeling | |
| Wang et al. | Adaafford: Learning to adapt manipulation affordance for 3d articulated objects via few-shot interactions | |
| Jaques et al. | Physics-as-inverse-graphics: Unsupervised physical parameter estimation from video | |
| Du et al. | Unsupervised adversarial domain adaptation for micro-Doppler based human activity classification | |
| Gu et al. | Multiple stream deep learning model for human action recognition | |
| Chang et al. | Fast Random‐Forest‐Based Human Pose Estimation Using a Multi‐scale and Cascade Approach | |
| CN113313123B (en) | Glance path prediction method based on semantic inference | |
| Yu et al. | Human motion based intent recognition using a deep dynamic neural model | |
| CN105740815B (en) | A kind of Human bodys' response method based on depth recurrence stratified condition random field | |
| CN107341471B (en) | A kind of Human bodys' response method based on Bilayer condition random field | |
| CN116524593A (en) | Dynamic gesture recognition method, system, equipment and medium | |
| CN113989943B (en) | Distillation loss-based human body motion increment identification method and device | |
| CN113408721A (en) | Neural network structure searching method, apparatus, computer device and storage medium | |
| CN117272168A (en) | Human body action recognition and prediction method based on motion time sequence feature coding | |
| Xie et al. | A pyramidal deep learning architecture for human action recognition | |
| CN116089874A (en) | Emotion recognition method and device based on ensemble learning and migration learning | |
| Liu et al. | Deepssm: Deep state-space model for 3d human motion prediction | |
| Xu et al. | Scene-perception graph convolutional networks for human action prediction | |
| Kiruba et al. | Deep learning for human action recognition survey |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20160706 Assignee: Nanjing Nanyou Information Industry Technology Research Institute Co. Ltd. Assignor: Nanjing Post & Telecommunication Univ. Contract record no.: X2019980001257 Denomination of invention: Human body behavior identification method based on deep recursive and hierarchical condition random fields Granted publication date: 20181218 License type: Common License Record date: 20191224 |
|
| EE01 | Entry into force of recordation of patent licensing contract | ||
| EE01 | Entry into force of recordation of patent licensing contract |
Application publication date: 20160706 Assignee: Jiangsu Tuoyou Information Intelligent Technology Research Institute Co.,Ltd. Assignor: NANJING University OF POSTS AND TELECOMMUNICATIONS Contract record no.: X2021320000043 Denomination of invention: A human behavior recognition method based on deep recursive hierarchical conditional random field Granted publication date: 20181218 License type: Common License Record date: 20210616 |
|
| EC01 | Cancellation of recordation of patent licensing contract | ||
| EC01 | Cancellation of recordation of patent licensing contract |
Assignee: NANJING NANYOU INSTITUTE OF INFORMATION TECHNOVATION Co.,Ltd. Assignor: NANJING University OF POSTS AND TELECOMMUNICATIONS Contract record no.: X2019980001257 Date of cancellation: 20220304 |