CN108805058A - Target object changes gesture recognition method, device and computer equipment - Google Patents
Target object changes gesture recognition method, device and computer equipment Download PDFInfo
- Publication number
- CN108805058A CN108805058A CN201810531141.4A CN201810531141A CN108805058A CN 108805058 A CN108805058 A CN 108805058A CN 201810531141 A CN201810531141 A CN 201810531141A CN 108805058 A CN108805058 A CN 108805058A
- Authority
- CN
- China
- Prior art keywords
- target object
- frame image
- image
- joint point
- point diagram
- 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
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
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Biology (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Biophysics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
This application involves a kind of target object variation gesture recognition method, device, computer equipment and storage mediums, first identify the initial joint point diagram of the target object in image, operating procedure later is also to be directed to the target object progress of image respectively, the information that can include to the different target object in image parallel is handled, therefore the variation posture that can quickly identify multiple target objects in video data, improves the speed of response of system.In addition, the method that above-described embodiment refers to is when identifying the variation posture of target object, it is corrected to obtain joint point diagram composition artis graphic sequence according to corresponding initial joint point diagram in adjacent frame image, it can ensure to obtain stable artis, and then overcome target object and be in dynamic case, and leading to image jitter problem, the joint point diagram of target object is clearly in each frame image ensured, therefore, the recongnition of objects result obtained using the method for the present embodiment is clearly accurate.
Description
Technical field
This application involves technical field of computer vision, more particularly to a kind of target object variation gesture recognition method,
Device, computer equipment and storage medium.
Background technology
With the development of society, being carried out by interaction platform interactive, it has also become the important amusement side in people's modern life
Formula.
The attitudes vibration identification of target object is to realize one of the basic technology of platform interdynamic, but traditional technology is in video
In multiple target object attitudes vibration identification when response it is slow, seriously affect the Interactive Experience of user.
Invention content
Based on this, it is necessary in view of the above technical problems, provide one kind and quickly identify multiple target object postures in video
Method, apparatus, computer equipment and the storage medium of variation.
In a first aspect, the embodiment of the present application provides a kind of target object variation gesture recognition method, include the following steps：
Video data is obtained, and the video data is decoded, obtains multiple image；
The multiple image is inputted into neural network model, obtains initial joint point of the target object in every frame image
Figure；
Using the initial joint point diagram of same target object in the adjacent frame image of every frame image, correct per corresponding in frame image
Initial joint point diagram, obtain it is described per frame image in target object joint point diagram；
According to each frame image when ordered pair described in each frame image the joint point diagram of target object be ranked up, obtain described
The artis graphic sequence of target object；
According to the artis graphic sequence of the target object, the variation posture of the target object is identified.
It is described in one of the embodiments, that the multiple image is inputted into neural network model, it obtains target object and exists
The step of initial joint point diagram in per frame image includes：
Using the convolutional neural networks of the neural network model, convolution behaviour is carried out to each frame image in the multiple image
Make, extracts the feature of multiple initial joint points and target object in each frame image；
According to the feature of the target object, sort operation is carried out to multiple initial joint points in each frame image respectively,
Obtain initial joint point diagram of the target object in every frame image.
The initial joint using same target object in the often adjacent frame image of frame image in one of the embodiments,
Point diagram is corrected and obtains the joint point diagram of target object in the image per frame per corresponding initial joint point diagram in frame image
Step includes：
Obtain the boundary rectangle frame of each target object in each frame image；
Calculate each boundary rectangle frame in each frame image and the friendship of each boundary rectangle frame and ratio in adjacent frame image；
According to the friendship and ratio and predetermined threshold value, each frame image and the same target object in adjacent frame image are determined；
It is original in corresponding each frame image of amendment using the initial joint point diagram of same target object in the adjacent frame image
Joint point diagram obtains the joint point diagram of target object in each frame image.
Described the step of obtaining the boundary rectangle frame of each target object in each frame image, wraps in one of the embodiments,
It includes：
All target objects in any one frame image are identified, according to each target object in all target objects
Initial joint point diagram generates the boundary rectangle frame of each target object in any one frame image.
If the target object is human body in one of the embodiments, the multiple image is inputted into neural network
Model, the step of obtaining initial joint point diagram of the target object in every frame image include：
The multiple image is inputted into neural network model, obtains multiple initial joint points and each frame figure in each frame image
The limbs distribution map of human body as in；
According to the limbs distribution map of human body in each frame image, multiple initial joints in each frame image are clicked through respectively
Row cluster, obtains personage's classification belonging to original artis in each frame image；
According to personage's classification belonging to the initial joint point, initial joint point of each human body in each frame image is obtained
Figure.
In one of the embodiments, before the step of multiple image is inputted neural network model, the side
Method further includes：
Training image is obtained, and the distribution of the limbs of artis and human body in the training image is labeled, is obtained
Multiple model training samples；
Using the multiple original neural network model of model training sample training, the neural network model is obtained.
It is described in one of the embodiments, to use the multiple original neural network model of model training sample training,
The step of obtaining the neural network model include：
The neural network in the neural network model is built according to goal task；
The multiple model training sample is inputted into the neural network, iteration executes the positive of the neural network and derives
And back-propagation algorithm, obtain the neural network model；
The goal task includes：Skeletal joint point returns task, limbs distribution recurrence task and personage's categorical clusters and appoints
Business.
Second aspect, the embodiment of the present application provide a kind of target object variation gesture recognition device, and described device includes：
Acquisition module is decoded for obtaining video data, and to the video data, obtains multiple image；
Artis identification module obtains target object in every frame for the multiple image to be inputted neural network model
Initial joint point diagram in image；
Artis processing module, for the initial joint point using same target object in the adjacent frame image per frame image
Figure corrects the joint point diagram that target object in the image per frame is obtained per corresponding initial joint point diagram in frame image；According to
Each frame image when ordered pair described in each frame image the joint point diagram of target object be ranked up, obtain the pass of the target object
Node graphic sequence；According to the artis graphic sequence of the target object, the variation posture of the target object is identified.
The third aspect, the embodiment of the present application provide a kind of computer equipment, including memory and processor, the memory
On be stored with the computer program that can be run on a processor, the processor realizes the application when executing the computer program
The image processing method that any embodiment provides.
Fourth aspect, the embodiment of the present application provide a kind of computer readable storage medium, are stored thereon with computer program,
The computer program is executed by processor the image processing method of the application any embodiment offer.
This application involves a kind of target object variation gesture recognition method, device, computer equipment and storage medium, the mesh
Mark object variation gesture recognition method identifies the initial joint point diagram of the target object in image, operating procedure later first
Be respectively be directed to image target object carry out, you can with include to the different target object in image parallel information at
Reason, therefore can quickly identify the variation posture of multiple target objects in video data, improve the speed of response of system.In addition,
The method that above-described embodiment refers to is when identifying the variation posture of target object, according to corresponding initial joint point in adjacent frame image
Figure, which is corrected, obtains joint point diagram composition artis graphic sequence, it is ensured that obtains stable artis, and then overcomes target pair
As being in dynamic case, and lead to image jitter problem, the joint point diagram of target object is clear in each frame image ensured
Clear, therefore, the recongnition of objects result obtained using the method for the present embodiment is clearly accurate.
Description of the drawings
Fig. 1 is the applied environment figure that target object changes gesture recognition method in the application one embodiment；
Fig. 2 is the flow diagram that target object changes gesture recognition method in the application one embodiment；
Fig. 3 is to be flowed the step of obtaining initial joint point diagram of the target object in every frame image in the application one embodiment
Journey schematic diagram；
Fig. 4 is the step flow signal for the joint point diagram that target object in multiple image is obtained in the application one embodiment
Figure；
Fig. 5 is the step of obtaining initial joint point diagram of the target object in every frame image in the application another embodiment
Flow diagram；
Fig. 6 is the supplement steps flow chart schematic diagram that target object changes gesture recognition method in the application one embodiment；
Fig. 7 is to obtain the step flow diagram of neural network model in the application one embodiment；
Fig. 8 is the structure diagram that target object changes gesture recognition device in the application one embodiment；
Fig. 9 is the internal structure chart of the application one embodiment Computer equipment.
Specific implementation mode
It is with reference to the accompanying drawings and embodiments, right in order to make the object, technical solution and advantage of the application be more clearly understood
The application is further elaborated.It should be appreciated that specific embodiment described herein is only used to explain the application, not
For limiting the application.
Target object provided by the present application changes gesture recognition method, can be applied in application environment as shown in Figure 1.
Wherein, video acquisition device 110 is connect with computer equipment 120.Wherein, video acquisition device 110 can with computer equipment 120
To be set as integral type terminal, which can be including but not limited to being various smart mobile phones, tablet computer and portable
Formula wearable device, personal computer, laptop.The video acquisition device 110 and computer equipment 120 can also be point
The video capture device and server of not independent setting.The video capture device is connect by network with server communication, should
Server can be realized with the server cluster of the either multiple server compositions of independent server.Optionally, the video
Collecting device can be mobile phone camera, monitoring camera, vehicle-mounted camera, one or more of common camera.
Optionally, which includes at least one processor 121 and memory 122.Optionally, the processing
Device 121 can be CPU (Central Processing Unit, central processing unit), GPU (Graphics Processing
Unit, graphics processor), IPU (Intelligence Processing Unit, intelligent processor) etc. is preferably suitable to figure
As the GPU or the IPU suitable for running neural network algorithm of processing.Optionally, which is multi-core processor, such as
Multinuclear GPU.
Optionally, the neural network algorithm of image procossing is stored on the memory 122.Image is carried out in processor 121
When processing, the neural network algorithm stored on memory 122 can be called to carry out image procossing.
Optionally, multiple processor cores of multiple processors 121 of the computer equipment or a certain processor 121 can be with
The multiple operation tasks of parallel processing, to improve the data-handling efficiency of computer equipment 120.It optionally it should be understood that should
Processor 121 can run the algorithm of neural network model, and the data to inputting neural network model are handled.Optionally,
The data for inputting neural network model, can be initial data, such as the photograph of acquisition, video etc.；Can also be encoded
Data with set form afterwards, such as the data etc. through one-hot codings.
In one embodiment, as shown in Fig. 2, a kind of target object variation gesture recognition method is provided, now with Fig. 1 institutes
It illustrates, includes the following steps for the application environment shown：
Step S210：Video data is obtained, and the video data is decoded, obtains multiple image.
Specifically, after the processor 121 of computer equipment 120 obtains video data, the video data is decoded,
Obtain multiple image.Optionally, which can be the video data for the rgb format that video acquisition device 110 acquires.
Step S220：The multiple image is inputted into neural network model, obtains original of the target object in every frame image
Beginning joint point diagram.
Wherein, joint point diagram is the image being made of artis.Initial joint point is after neural network model handles image
The artis directly exported.Initial joint point diagram in the present embodiment refers to the artis group directly exported by neural network model
At image.Specifically, the processor 121 of computer equipment 120 calls phase after obtaining multiple image from memory 122
The neural network algorithm of pass handles the multiple image of acquisition, obtains initial joint of the target object in every frame image
Point diagram.Optionally, processor 121 runs the calculation of each network layer of neural network model after obtaining the decoded multiple image of video
Method handles the multiple image, handles each frame image of input, obtains initial joint of the target object in every frame image
Point diagram.
Optionally, processor 121 can identify multiple original passes in current frame image first when handling multiple image
Node determines these initial joint points belong to which target object in current frame image again later, that is, determines each target object
Initial joint point diagram be made of which initial joint point.Similarly, processor 121 handles other frame images, obtains target object
Initial joint point diagram in every frame image.
Such as：After a certain video data decoding, N frame images are obtained, include target object in the video data：Human body X
With human body Y.Processor 121 obtains in first frame image first when operation neural network model handles the N frame images
Including multiple initial joint points, result is：A, b, c, d, e, f determine multiple initial joint point a, b, c, d, e, f difference later
Which target object belonged to, the initial joint point diagram and target object human body Y of target object human body X in first frame image are obtained
Initial joint point diagram, wherein the initial joint point diagram of human body X includes initial joint point a, b, c in first frame image, human body Y's
Initial joint point diagram includes initial joint point d, e, f.The second frame image is similarly handled to nth frame image, obtains the second frame image
To the initial joint point diagram of the initial joint point diagram and target object human body Y of target object human body X in nth frame image.
Step S230：Using the initial joint point diagram of same target object in the adjacent frame image of every frame image, correct per frame
Corresponding initial joint point diagram in image obtains the joint point diagram of target object in the image per frame.
Specifically, processor 121 is repaiied using the initial joint point diagram of same target object in the adjacent frame image per frame image
Corresponding initial joint point diagram in just every frame image, obtains the joint point diagram of target object in the image per frame.Optionally, locate
Reason device 121 identify current frame image and the same target object in adjacent frame image first, the use of preprocessor 121 neighbour's frame figure
As in the initial joint point diagram that belongs to same target object in current frame image, the corresponding original pass corrected in current frame image
Node diagram.Optionally, the position belonging to initial joint point diagram of the processor 121 first according to same target object determines neighbour's frame
Image initial joint point corresponding with current frame image position, it is corresponding again later to adjust each original key point in current frame image
Position is to get to the joint point diagram by target object in initial joint point forms after adjusting current frame image.Optionally, each frame
The adjacent frame image of image can be a frame or multiple image adjacent with target frame image.Optionally, the adjacent frame figure of each frame image
As that can be the previous frame image of target frame figure or a later frame image of target frame figure.
Such as：In obtained N frame images, the initial joint point diagram A1 of the target object human body X of first frame image, target
The initial joint point diagram D1 of object human body Y；Target object human body X includes initial joint point diagram A2, target pair in second frame image
As human body Y includes initial joint point diagram D2；……；It is An that target object human body X, which includes initial joint point diagram, in nth frame image,
Target object human body Y includes initial joint point diagram Dn.Processor 121 is in the original pass according to same target object in adjacent frame image
Node diagram identifies the human body X in first frame image and second first when correcting corresponding initial joint point diagram in each frame image
Human body X in frame image, the human body Y in human body Y and the second frame image in first frame image, reuses human body X later
Initial joint point diagram A1 in one frame image corrects the initial joint point diagram A2 of human body X in the second frame image, obtains the second frame figure
The joint point diagram A2 ' of target object human body X as in corrects the using initial joint point diagram D1 of the human body Y in first frame image
The initial joint point diagram D2 of human body Y in two frame images, obtains the joint point diagram D2 ' of target object human body Y in the second frame image.Together
Reason processing third frame image obtains joint the point diagram A3's ' and human body Y of third frame image object object human body X to nth frame image
Joint point diagram D3 ' ... ..., the joint point diagram Dn ' of joint the point diagram An ' and human body Y of target object human body X in nth frame image.
It should be noted that the selection of adjacent frame image can be selected according to actual demand, the purpose of this step is intended to
According to the artis information in each frame image of artis Information revision in adjacent frame image, and then obtain the pass of stable target object
Node diagram does not limit the selection mode of adjacent frame image at this.
Step S240：According to each frame image when ordered pair described in each frame image the joint point diagram of target object arranged
Sequence obtains the artis graphic sequence of the target object.
Wherein, artis graphic sequence refers to the pass man point set of graphs arranged according to certain sequential.Specifically, processor 121
According to each frame image when ordered pair described in each frame image the joint point diagram of target object be ranked up, obtain the target object
Artis graphic sequence.
Step S250：According to the artis graphic sequence of the target object, the variation posture of the target object is identified.
Specifically, processor 121 identifies the variation of the target object according to the artis graphic sequence of the target object
Posture.Optionally, processor 121 is smoothed obtained artis graphic sequence, obtains the variation posture of target object
Recognition result.
A kind of target object in above-described embodiment changes gesture recognition method, first identifies the original of the target object in image
Beginning joint point diagram, operating procedure later is also to be directed to the target object progress of image respectively, you can with parallel in image
The information that different target object includes is handled, therefore can quickly identify the variation appearance of multiple target objects in video data
State improves the speed of response of system.In addition, the method that refers to of above-described embodiment is when identifying the variation posture of target object, root
It corrects to obtain joint point diagram composition artis graphic sequence according to corresponding initial joint point diagram in adjacent frame image, it is ensured that obtain steady
Fixed artis information, and then when overcoming target object and being in dynamic case, lead to image jitter problem, what is ensured is each
In frame image the joint point diagram of target object be clearly, therefore, the recongnition of objects obtained using the method for the present embodiment
As a result clear accurate.
As an optional implementation manner, as shown in figure 3, described input neural network model by the multiple image,
The step of obtaining initial joint point diagram of the target object in every frame image include：
Step S221：Using the convolutional neural networks of the neural network model, to each frame image in the multiple image
Convolution operation is carried out, the feature of multiple initial joint points and target object in each frame image is extracted.
Wherein, the feature of target object refers to the feature that can be used for distinguishing current target object and other target objects.
Specifically, multiple image is inputted neural network model by the processor 121 of computer equipment 120 after obtaining multiple image, right
Each frame image carries out convolution operation in the multiple image, obtains multiple initial joint points in each frame image and target object
Feature.Optionally, the feature of the target object can be consistent in each frame image, can also be all one group corresponding per frame image
The feature of target object.Optionally, after model data is inputted neural network model by processor 121, the neural network mould is run
The algorithm of each network layer of type carries out calculation process to the model data of input, and final output obtains multiple in described image
The feature of target object in artis and described image.
Step S222：According to the feature of the target object, multiple initial joint points in each frame image are carried out respectively
Sort operation obtains initial joint point diagram of the target object in every frame image.
Specifically, the processor 121 of computer equipment 120 uses the grader of neural network according to the target object
Feature carries out sort operation to multiple initial joint points in each frame image respectively, obtains target object in every frame image
Initial joint point diagram.Optionally, processor 121 obtains the feature of target object in image first, further according to the spy of target object
Sign classifies to the artis of acquisition, obtains the target object classification belonging to each artis to get to the pass of each target object
Node.
The present embodiment carries out feature extraction, and the feature of the target object according to extraction using neural network model to image
Classify to the initial joint point of extraction, this is the feature extraction and classifying carried out based on image unique characteristics, therefore is directed to
Property is strong, and obtained result is accurate.
As an optional implementation manner, as shown in figure 4, described use same target in the often adjacent frame image of frame image
The initial joint point diagram of object is corrected per corresponding initial joint point diagram in frame image, and target pair in the image per frame is obtained
The step of joint point diagram of elephant includes：
Step S231：Obtain the boundary rectangle frame of each target object in each frame image.
Specifically, processor 121 obtains the boundary rectangle frame of each target object in each frame image.Optionally, processor
121 identify the target object in each frame image first, generate boundary rectangle frame based on the target object identified later.
Step S232：Calculate the friendship of each boundary rectangle frame and each boundary rectangle frame in adjacent frame image in each frame image
And compare.
Specifically, processor 121 calculates each boundary rectangle frame in each frame image and each external square in adjacent frame image
The friendship of shape frame and than (Intersection-Over-Union, IOU).Optionally, processor 121 calculates each frame image first
In include boundary rectangle frame area, calculate later each boundary rectangle frame of each frame image with it is each external in adjacent frame image
The overlapping area of rectangle frame, it is last to be obtained in each frame image according to obtained overlapping area and corresponding boundary rectangle frame area
Each boundary rectangle frame and the friendship of each boundary rectangle frame and ratio in adjacent frame image.Optionally, it is calculated using following formula external
The friendship of rectangle frame and ratio：Area/(the rectangle frame A areas+rectangle frame B areas) of friendship and ratio=2* boundary rectangle frames overlapping.Such as：
Include boundary rectangle a frame X, area A3 in third frame image；Include a boundary rectangle frame in the 4th frame image
Y, area A4；The overlapping area of two boundary rectangle frames is Ax, then, the friendship of boundary rectangle frame X and boundary rectangle frame is simultaneously
Than for：Friendship and ratio=2*Ax/ (A3+A4).
Step S233：According to the friendship and ratio and predetermined threshold value, each frame image and the same target in adjacent frame image are determined
Object.
Specifically, processor 121 is according to the friendship and ratio and predetermined threshold value, determine each frame image with it is same in neighbour's frame image
One target object.Optionally, confidence is matched according to the attitudes vibration speed degree of target object and/or video capture device 110
Breath is arranged the friendship and compares.It is alternatively possible to when being preset in friendship and comparing more than predetermined threshold value, by the corresponding mesh of two boundary rectangle frames
Mark object is determined as same target object.
Step S234：It is corresponding to correct each frame figure using the initial joint point diagram of same target object in the adjacent frame image
Initial joint point diagram as in, obtains the joint point diagram of target object in each frame image.
Specifically, processor 121 uses the initial joint point diagram of same target object in the adjacent frame image, corresponding amendment
Initial joint point diagram in each frame image obtains the joint point diagram of target object in each frame image.Optionally, processor 121 is first
It first determines current frame image and the same target object in adjacent frame image, is determining same target object in the two field pictures later
In include in original artis figure initial joint point correspondence, finally according to the correspondence of original artis in two field pictures
Relationship, correct current frame image in target object initial joint figure in original artis, obtain target pair in current frame image
The joint point diagram of elephant.
The present embodiment by the boundary rectangle frame of the target object of generation to position of the target object in each frame image into
Row tracking, can fast and accurately determine the same target object in different frame image.
As an optional implementation manner, the step for obtaining the boundary rectangle frame of each target object in each frame image
Suddenly include：All target objects in any one frame image are identified, according to each target object in all target objects
Initial joint point diagram generates the boundary rectangle frame of each target object in any one frame.
Optionally, processor 121 identifies all target objects in any one frame image, according to all target objects
In each target object initial joint point diagram, generate the boundary rectangle frame of each target object in any one frame image.
Optionally, processor 121 can be based on faster-RCNN neural networks and build neural network model, to obtained each frame image
It is handled, obtains the boundary rectangle frame of each target object.Since faster-RCNN neural networks are generating boundary rectangle frame
When, data processing is efficient, and the initial boundary rectangle frame that the faster-RCN meeting refine of the neural network model obtains, therefore can
Boundary rectangle frame to ensure obtained target object is accurate and reliable.
As an optional implementation manner, above-mentioned target object, which can be human body, animal, robot etc., has artis
Object.It should be understood that the feature for the target object that data processing obtains is carried out using neural network model, for corresponding mesh
Mark the feature of object.Such as target object, when being human body, the feature of the target object can be the limbs distribution map of human body, the limb
The distributional pattern of partes corporis humani position in body distribution table diagram picture.
As an optional implementation manner, as shown in figure 5, when target object is human body, the multiple image is defeated
The step of entering neural network model, obtaining initial joint point diagram of the target object in every frame image include：
Step S221 '：The multiple image is inputted into neural network model, obtains multiple initial joints in each frame image
The limbs distribution map of human body in point and each frame image.
Step S222 '：According to the limbs distribution map of human body in each frame image, respectively to multiple originals in each frame image
Beginning artis is clustered, and personage's classification belonging to original artis in each frame image is obtained.
Step S223 '：According to personage's classification belonging to the initial joint point, each human body is obtained in each frame image
Initial joint point diagram.
Wherein, personage's classification is the classification divided according to different human body in image.Specifically, processor 121 is run first
Each layer network of neural network model is handled to obtain multiple initial joint points in image to the image of input model, and
The limbs distribution map of human body.Later according to the limbs distribution map of the human body in image, multiple initial joints in image are clicked through
Row cluster, obtains personage's classification belonging to each artis in image.Optionally, processor 121 determines human body in image first
Number, and based on the number setting other classification number of figure kind to get to the artis that each human body includes in image.
The method that the initial joint point diagram of each human body is obtained in above-described embodiment is primarily based on neural network model proposition
The limbs distribution characteristics of initial joint point and human body in image, later according to the limbs distribution characteristics of human body to initial joint point
It is clustered, can fast and accurately identify which human body the initial joint point in image belongs to.
As an optional implementation manner, as shown in fig. 6, in the step that the multiple image is inputted to neural network model
Before rapid, the method further includes：
Step S260：Training image is obtained, and the limbs of artis and human body in the training image are distributed and are carried out
Mark, obtains multiple model training samples.
Specifically, using annotation tool (such as the softwares such as Photoshop) to manually being marked in training image.It is optional
Ground includes to carrying out artificial mark in training image：Mark the artis of each human body in training image；And according to human body
Physiological characteristic marks the artis line of each human body, that is, marks the limbs distribution map of human body.It optionally, can also be to the instruction
Practice image and carry out data enhancing pretreatment, such as training image is overturn and/or disturbed operation.
Step S270：Using the multiple original neural network model of model training sample training, the nerve net is obtained
Network model.
Specifically, the neural network structure for neural network model being built previously according to goal task uses step later
Multiple original neural network models of model training sample training that S260 is obtained, until the network convergence of network model, obtains god
Through network model.The goal task includes：Skeletal joint point returns task, limbs distribution recurrence task and personage's categorical clusters and appoints
Business.
The neural network model that the present embodiment is trained, may be implemented the artis identification of human body in image, and obtain
The limbs distribution characteristics of each human body in image later again clicks through the joint identified based on the limbs distribution characteristics for obtaining human body
Pedestrian's object category classification, thus the model can based on image itself comprising feature (non-universal category feature) in image
Target object is handled, therefore handling result is fine.
As an optional implementation manner, as shown in fig. 7, it is described original using the multiple model training sample training
Neural network model, the step of obtaining the neural network model include：
Step S271：The neural network in the neural network model is built according to goal task；The knot of the neural network
Structure includes that skeletal joint point returns task neural net layer, and limbs are distributed recurrence task neural net layer and personage's categorical clusters are appointed
Business neural net layer.
Step S272：The multiple model training sample is inputted into the neural network, iteration executes the neural network
It is positive derive and back-propagation algorithm, obtain the neural network model.Optionally, which includes：
The multiple model training spurious edition is input to the neural network by step a, using stochastic gradient descent algorithm,
The Grad of the loss function of each layer of the neural network is calculated, and uses obtained each layer of Grad update respective layer
Weights；Step b calculates the error-sensitivity of each layer of the neural network, and use each layer of mistake using back-propagation algorithm
The weights of poor sensitivity update respective layer；Step c, iteration execute the step a and step b, until using each layer of gradient
The weights of value update respective layer are equal with the error-sensitivity update weights of respective layer using each layer, complete neural network
Training.
The model training method that the present embodiment proposes, obtained neural network model, when carrying out data processing, the place of output
It is accurate and reliable to manage result.
It should be understood that although each step in the flow chart of Fig. 2-Fig. 7 is shown successively according to the instruction of arrow,
Be these steps it is not that the inevitable sequence indicated according to arrow executes successively.Unless expressly stating otherwise herein, these steps
There is no stringent sequences to limit for rapid execution, these steps can execute in other order.Moreover, in Fig. 2-Fig. 7 extremely
Few a part of step may include that either these sub-steps of multiple stages or stage are not necessarily same to multiple sub-steps
Moment executes completion, but can execute at different times, and the execution sequence in these sub-steps or stage is also not necessarily
It carries out successively, but can either the sub-step of other steps or at least part in stage in turn or are handed over other steps
Alternately execute.
In one embodiment, as shown in figure 8, providing a kind of target object variation gesture recognition device, described device
Including：
Acquisition module 810 is decoded for obtaining video data, and to the video data, obtains multiple image.
Artis identification module 820 obtains target object every for the multiple image to be inputted neural network model
Initial joint point diagram in frame image.
Artis processing module 830, for the initial joint using same target object in the adjacent frame image per frame image
Point diagram corrects the joint point diagram that target object in the image per frame is obtained per corresponding initial joint point diagram in frame image；It presses
According to each frame image when ordered pair described in each frame image the joint point diagram of target object be ranked up, obtain the target object
Artis graphic sequence；According to the artis graphic sequence of the target object, the variation posture of the target object is identified.
As an optional implementation manner, artis identification module 820 is specifically used for using the neural network model
Convolutional neural networks, convolution operation is carried out to each frame image in the multiple image, extracts multiple original passes in each frame image
The feature of node and target object；According to the feature of the target object, respectively to multiple initial joint points in each frame image
Sort operation is carried out, initial joint point diagram of the target object in every frame image is obtained.
As an optional implementation manner, artis processing module 830 is specifically used for obtaining each mesh in each frame image
Mark the boundary rectangle frame of object；Calculate each boundary rectangle frame in each frame image and each boundary rectangle frame in adjacent frame image
It hands over and compares；According to the friendship and ratio and predetermined threshold value, each frame image and the same target object in adjacent frame image are determined；Using institute
The initial joint point diagram of same target object in adjacent frame image is stated, the corresponding initial joint point diagram corrected in each frame image obtains
The joint point diagram of target object in each frame image.
As an optional implementation manner, artis processing module 830, it is all in any one frame image for identification
Target object generates any one frame figure according to the initial joint point diagram of each target object in all target objects
The boundary rectangle frame of each target object as in.
As an optional implementation manner, artis identification module 820, for the multiple image to be inputted nerve net
Network model obtains the limbs distribution map of human body in multiple initial joint points and each frame image in each frame image；According to each frame figure
The limbs distribution map of human body, respectively clusters multiple initial joint points in each frame image, obtains each frame figure as in
Personage's classification as in belonging to original artis；According to personage's classification belonging to the initial joint point, obtains each human body and exist
Initial joint point diagram in each frame image.
As an optional implementation manner, artis identification module 820 uses the multiple model training sample training
Original neural network model obtains the neural network model.
As an optional implementation manner, artis identification module 820, for building the nerve according to goal task
Neural network in network model；The multiple model training sample is inputted into the neural network, iteration executes the nerve
The positive derivation of network and back-propagation algorithm, obtain the neural network model；The goal task includes：Skeletal joint point
Recurrence task, limbs distribution recurrence task and personage's categorical clusters task.
The specific restriction for changing gesture recognition device about target object may refer to change above for target object
The restriction of gesture recognition method, details are not described herein.Modules in above-mentioned target object variation gesture recognition device can be complete
Portion or part are realized by software, hardware and combinations thereof.Above-mentioned each module can be in the form of hardware embedded in or independently of calculating
In processor in machine equipment, can also in a software form it be stored in the memory in computer equipment, in order to processor
It calls and executes the corresponding operation of the above modules.
In one embodiment, a kind of computer equipment is provided, which can be terminal, structural representation
Figure can be as shown in Figure 9.The computer equipment includes the processor connected by system bus, memory, network interface, display
Screen and input unit.Wherein, the processor of the computer equipment is for providing calculating and control ability.The computer equipment is deposited
Reservoir includes non-volatile memory medium, built-in storage.The non-volatile memory medium is stored with operating system and computer journey
Sequence.The built-in storage provides environment for the operation of operating system and computer program in non-volatile memory medium.The calculating
The network interface of machine equipment is used to communicate by network connection with external terminal.When the computer program is executed by processor with
Realize a kind of target object variation gesture recognition method.The display screen of the computer equipment can be liquid crystal display or electronics
The input unit of ink display screen, the computer equipment can be the touch layer covered on display screen, can also be that computer is set
Button, trace ball or the Trackpad being arranged on standby shell, can also be external keyboard, Trackpad or mouse etc..
It will be understood by those skilled in the art that structure shown in Fig. 9, is only tied with the relevant part of application scheme
The block diagram of structure does not constitute the restriction for the computer equipment being applied thereon to application scheme, specifically computer equipment
May include either combining certain components than more or fewer components as shown in the figure or being arranged with different components.
In one embodiment, a kind of computer equipment, including memory and processor are provided, is deposited on the memory
The computer program that can be run on a processor is contained, processor realizes following steps when executing computer program：Obtain video
Data, and the video data is decoded, obtain multiple image；The multiple image is inputted into neural network model, is obtained
To initial joint point diagram of the target object in every frame image；Use the original of same target object in the adjacent frame image of every frame image
Beginning joint point diagram corrects the joint that target object in the image per frame is obtained per corresponding initial joint point diagram in frame image
Point diagram；According to each frame image when ordered pair described in each frame image the joint point diagram of target object be ranked up, obtain the mesh
Mark the artis graphic sequence of object；According to the artis graphic sequence of the target object, the variation appearance of the target object is identified
State.
In one embodiment, following steps are also realized when processor executes computer program：Use the neural network
The convolutional neural networks of model carry out convolution operation to each frame image in the multiple image, extract multiple originals in each frame image
The feature of beginning artis and target object；According to the feature of the target object, respectively to multiple original passes in each frame image
Node carries out sort operation, obtains initial joint point diagram of the target object in every frame image.
In one embodiment, following steps are also realized when processor executes computer program：It obtains every in each frame image
The boundary rectangle frame of one target object；Calculate each boundary rectangle frame in each frame image and each boundary rectangle in adjacent frame image
The friendship of frame and ratio；According to the friendship and ratio and predetermined threshold value, each frame image and the same target object in adjacent frame image are determined；Make
With the initial joint point diagram of same target object in the adjacent frame image, the initial joint point diagram corrected in each frame image is corresponded to,
Obtain the joint point diagram of target object in each frame image.
In one embodiment, following steps are also realized when processor executes computer program：Identify any one frame image
In all target objects, according to the initial joint point diagram of each target object in all target objects, generate it is described certain
The boundary rectangle frame of each target object in frame image.
In one embodiment, when target object is human body in the image of processing, when processor executes computer program
Also realize following steps：The multiple image is inputted into neural network model, obtains multiple initial joint points in each frame image
With the limbs distribution map of human body in each frame image；According to the limbs distribution map of human body in each frame image, respectively to each frame figure
Multiple initial joint points as in are clustered, and personage's classification belonging to original artis in each frame image is obtained；According to described
Personage's classification belonging to initial joint point obtains initial joint point diagram of each human body in each frame image.
In one embodiment, following steps are also realized when processor executes computer program：Use the multiple model
Training sample trains original neural network model, obtains the neural network model.
In one embodiment, following steps are also realized when processor executes computer program：It is built according to goal task
Neural network in the neural network model；The multiple model training sample is inputted into the neural network, iteration executes
The positive derivation of the neural network and back-propagation algorithm, obtain the neural network model；The goal task includes：Bone
Bone artis returns task, limbs distribution recurrence task and personage's categorical clusters task.
In one embodiment, a kind of computer readable storage medium is provided, computer program is stored thereon with, is calculated
Machine program realizes the image processing method that the application any embodiment provides when being executed by processor.
One of ordinary skill in the art will appreciate that realizing all or part of flow in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the computer program can be stored in a non-volatile computer
In read/write memory medium, the computer program is when being executed, it may include such as the flow of the embodiment of above-mentioned each method.Wherein,
Any reference to memory, storage, database or other media used in each embodiment provided herein,
Including non-volatile and/or volatile memory.Nonvolatile memory may include read-only memory (ROM), programming ROM
(PROM), electrically programmable ROM (EPROM), electrically erasable ROM (EEPROM) or flash memory.Volatile memory may include
Random access memory (RAM) or external cache.By way of illustration and not limitation, RAM is available in many forms,
Such as static state RAM (SRAM), dynamic ram (DRAM), synchronous dram (SDRAM), double data rate sdram (DDRSDRAM), enhancing
Type SDRAM (ESDRAM), synchronization link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM
(RDRAM), direct memory bus dynamic ram (DRDRAM) and memory bus dynamic ram (RDRAM) etc..
Each technical characteristic of above example can be combined arbitrarily, to keep description succinct, not to above-described embodiment
In each technical characteristic it is all possible combination be all described, as long as however, the combination of these technical characteristics be not present lance
Shield is all considered to be the range of this specification record.
The several embodiments of the application above described embodiment only expresses, the description thereof is more specific and detailed, but simultaneously
It cannot therefore be construed as limiting the scope of the patent.It should be pointed out that coming for those of ordinary skill in the art
It says, under the premise of not departing from the application design, various modifications and improvements can be made, these belong to the protection of the application
Range.Therefore, the protection domain of the application patent should be determined by the appended claims.
Claims (10)
1. a kind of target object changes gesture recognition method, which is characterized in that include the following steps：
Video data is obtained, and the video data is decoded, obtains multiple image；
The multiple image is inputted into neural network model, obtains initial joint point diagram of the target object in every frame image；
Using the initial joint point diagram of same target object in the adjacent frame image of every frame image, correct per corresponding original in frame image
Beginning joint point diagram obtains the joint point diagram of target object in the image per frame；
According to each frame image when ordered pair described in each frame image the joint point diagram of target object be ranked up, obtain the target
The artis graphic sequence of object；
According to the artis graphic sequence of the target object, the variation posture of the target object is identified.
2. according to the method described in claim 1, it is characterized in that, it is described by the multiple image input neural network model,
The step of obtaining initial joint point diagram of the target object in every frame image include：
Using the convolutional neural networks of the neural network model, convolution operation is carried out to each frame image in the multiple image,
Extract the feature of multiple initial joint points and target object in each frame image；
According to the feature of the target object, sort operation is carried out to multiple initial joint points in each frame image respectively, is obtained
Initial joint point diagram of the target object in every frame image.
3. method according to claim 1 or 2, which is characterized in that same in the adjacent frame image using per frame image
The initial joint point diagram of target object is corrected per corresponding initial joint point diagram in frame image, and mesh in the image per frame is obtained
Mark object joint point diagram the step of include：
Obtain the boundary rectangle frame of each target object in each frame image；
Calculate each boundary rectangle frame in each frame image and the friendship of each boundary rectangle frame and ratio in adjacent frame image；
According to the friendship and ratio and predetermined threshold value, each frame image and the same target object in adjacent frame image are determined；
Use the initial joint point diagram of same target object in the adjacent frame image, the corresponding initial joint corrected in each frame image
Point diagram obtains the joint point diagram of target object in each frame image.
4. according to the method described in claim 3, it is characterized in that, described obtain the external of each target object in each frame image
The step of rectangle frame includes：
Identify all target objects in any one frame image, according in all target objects each target object it is original
Joint point diagram generates the boundary rectangle frame of each target object in any one frame image.
5. if according to the method described in claim 1, it is characterized in that, the target object is human body, by the multiframe figure
As the step of inputting neural network model, obtaining initial joint point diagram of the target object in every frame image includes：
The multiple image is inputted into neural network model, is obtained in multiple initial joint points and each frame image in each frame image
The limbs distribution map of human body；
According to the limbs distribution map of human body in each frame image, multiple initial joint points in each frame image are gathered respectively
Class obtains personage's classification belonging to original artis in each frame image；
According to personage's classification belonging to the initial joint point, initial joint point diagram of each human body in each frame image is obtained.
6. according to the method described in claim 5, it is characterized in that, in the step that the multiple image is inputted to neural network model
Before rapid, the method further includes：
Training image is obtained, and the distribution of the limbs of artis and human body in the training image is labeled, is obtained multiple
Model training sample；
Using the multiple original neural network model of model training sample training, the neural network model is obtained.
7. according to the method described in claim 6, it is characterized in that, described original using the multiple model training sample training
Neural network model, the step of obtaining the neural network model include：
The neural network in the neural network model is built according to goal task；
The multiple model training sample is inputted into the neural network, iteration executes the positive of the neural network and derives and anti-
To propagation algorithm, the neural network model is obtained；
The goal task includes：Skeletal joint point returns task, limbs distribution recurrence task and personage's categorical clusters task.
8. a kind of target object changes gesture recognition device, which is characterized in that described device includes：
Acquisition module is decoded for obtaining video data, and to the video data, obtains multiple image；
Artis identification module obtains target object in every frame image for the multiple image to be inputted neural network model
In initial joint point diagram；
Artis processing module is repaiied for the initial joint point diagram using same target object in the adjacent frame image per frame image
Corresponding initial joint point diagram in just every frame image, obtains the joint point diagram of target object in the multiple image；According to each frame
Image when ordered pair described in each frame image the joint point diagram of target object be ranked up, obtain the artis of the target object
Graphic sequence；According to the artis graphic sequence of the target object, the variation posture of the target object is identified.
9. a kind of computer equipment, including memory and processor, be stored on the memory to run on a processor
Computer program, which is characterized in that the processor realizes any one of claim 1 to 7 when executing the computer program
The step of the method.
10. a kind of computer readable storage medium, is stored thereon with computer program, which is characterized in that the computer program
The step of method described in any one of claim 1 to 7 is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810531141.4A CN108805058B (en) | 2018-05-29 | 2018-05-29 | Target object change posture recognition method and device and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810531141.4A CN108805058B (en) | 2018-05-29 | 2018-05-29 | Target object change posture recognition method and device and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108805058A true CN108805058A (en) | 2018-11-13 |
| CN108805058B CN108805058B (en) | 2020-12-15 |
Family
ID=64090740
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810531141.4A Active CN108805058B (en) | 2018-05-29 | 2018-05-29 | Target object change posture recognition method and device and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108805058B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110852237A (en) * | 2019-11-05 | 2020-02-28 | 浙江大华技术股份有限公司 | Object posture determining method and device, storage medium and electronic device |
| CN111126157A (en) * | 2019-11-27 | 2020-05-08 | 北京华捷艾米科技有限公司 | Data labeling method and device |
| WO2020107833A1 (en) * | 2018-11-26 | 2020-06-04 | 平安科技（深圳）有限公司 | Skeleton-based behavior detection method, terminal device, and computer storage medium |
| CN111241940A (en) * | 2019-12-31 | 2020-06-05 | 浙江大学 | Remote control method of robot and human body boundary frame determination method and system |
| CN111292402A (en) * | 2020-02-13 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Data processing method, device, equipment and computer readable storage medium |
| CN111310590A (en) * | 2020-01-20 | 2020-06-19 | 北京西米兄弟未来科技有限公司 | Action recognition method and electronic equipment |
| CN112488071A (en) * | 2020-12-21 | 2021-03-12 | 重庆紫光华山智安科技有限公司 | Method, device, electronic equipment and storage medium for extracting pedestrian features |
| CN112651291A (en) * | 2020-10-01 | 2021-04-13 | 新加坡依图有限责任公司(私有) | Video-based posture estimation method, device, medium and electronic equipment |
| CN113191368A (en) * | 2020-01-14 | 2021-07-30 | 北京地平线机器人技术研发有限公司 | Matching method and device of markers |
| CN113923493A (en) * | 2021-09-29 | 2022-01-11 | 北京奇艺世纪科技有限公司 | Video processing method and device, electronic equipment and storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7746345B2 (en) * | 1997-10-15 | 2010-06-29 | Hunter Kevin L | System and method for generating an animatable character |
| CN107644190A (en) * | 2016-07-20 | 2018-01-30 | 北京旷视科技有限公司 | Pedestrian's monitoring method and device |
| CN107920257A (en) * | 2017-12-01 | 2018-04-17 | 北京奇虎科技有限公司 | Video Key point real-time processing method, device and computing device |
-
2018
- 2018-05-29 CN CN201810531141.4A patent/CN108805058B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7746345B2 (en) * | 1997-10-15 | 2010-06-29 | Hunter Kevin L | System and method for generating an animatable character |
| CN107644190A (en) * | 2016-07-20 | 2018-01-30 | 北京旷视科技有限公司 | Pedestrian's monitoring method and device |
| CN107920257A (en) * | 2017-12-01 | 2018-04-17 | 北京奇虎科技有限公司 | Video Key point real-time processing method, device and computing device |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020107833A1 (en) * | 2018-11-26 | 2020-06-04 | 平安科技（深圳）有限公司 | Skeleton-based behavior detection method, terminal device, and computer storage medium |
| CN110852237A (en) * | 2019-11-05 | 2020-02-28 | 浙江大华技术股份有限公司 | Object posture determining method and device, storage medium and electronic device |
| CN111126157A (en) * | 2019-11-27 | 2020-05-08 | 北京华捷艾米科技有限公司 | Data labeling method and device |
| CN111126157B (en) * | 2019-11-27 | 2023-08-25 | 北京华捷艾米科技有限公司 | Data labeling method and device |
| CN111241940A (en) * | 2019-12-31 | 2020-06-05 | 浙江大学 | Remote control method of robot and human body boundary frame determination method and system |
| CN111241940B (en) * | 2019-12-31 | 2022-07-26 | 浙江大学 | Remote control method of robot and human body boundary frame determination method and system |
| CN113191368B (en) * | 2020-01-14 | 2023-08-01 | 北京地平线机器人技术研发有限公司 | Method and device for matching markers |
| CN113191368A (en) * | 2020-01-14 | 2021-07-30 | 北京地平线机器人技术研发有限公司 | Matching method and device of markers |
| CN111310590A (en) * | 2020-01-20 | 2020-06-19 | 北京西米兄弟未来科技有限公司 | Action recognition method and electronic equipment |
| CN111292402A (en) * | 2020-02-13 | 2020-06-16 | 腾讯科技（深圳）有限公司 | Data processing method, device, equipment and computer readable storage medium |
| CN111292402B (en) * | 2020-02-13 | 2023-03-07 | 腾讯科技（深圳）有限公司 | Data processing method, device, equipment and computer readable storage medium |
| CN112651291A (en) * | 2020-10-01 | 2021-04-13 | 新加坡依图有限责任公司(私有) | Video-based posture estimation method, device, medium and electronic equipment |
| CN112488071A (en) * | 2020-12-21 | 2021-03-12 | 重庆紫光华山智安科技有限公司 | Method, device, electronic equipment and storage medium for extracting pedestrian features |
| CN113923493B (en) * | 2021-09-29 | 2023-06-16 | 北京奇艺世纪科技有限公司 | Video processing method and device, electronic equipment and storage medium |
| CN113923493A (en) * | 2021-09-29 | 2022-01-11 | 北京奇艺世纪科技有限公司 | Video processing method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108805058B (en) | 2020-12-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108805058A (en) | Target object changes gesture recognition method, device and computer equipment | |
| CN110555481B (en) | Portrait style recognition method, device and computer readable storage medium | |
| CN109815826B (en) | Method and device for generating face attribute model | |
| CN109978754A (en) | Image processing method, device, storage medium and electronic equipment | |
| CN110378372A (en) | Diagram data recognition methods, device, computer equipment and storage medium | |
| US11417095B2 (en) | Image recognition method and apparatus, electronic device, and readable storage medium using an update on body extraction parameter and alignment parameter | |
| CN108230240A (en) | It is a kind of that the method for position and posture in image city scope is obtained based on deep learning | |
| CN109063584B (en) | Facial feature point positioning method, device, equipment and medium based on cascade regression | |
| CN106326857A (en) | Gender identification method and gender identification device based on face image | |
| CN109117760A (en) | Image processing method, device, electronic equipment and computer-readable medium | |
| CN114450690A (en) | Network model training method, image attribute identification method and device and electronic equipment | |
| CN112446302A (en) | Human body posture detection method and system, electronic equipment and storage medium | |
| CN108764143A (en) | Image processing method, device, computer equipment and storage medium | |
| CN108830782A (en) | Image processing method, device, computer equipment and storage medium | |
| US20200327726A1 (en) | Method of Generating 3D Facial Model for an Avatar and Related Device | |
| CN110163864A (en) | Image partition method, device, computer equipment and storage medium | |
| CN110390259A (en) | Recognition methods, device, computer equipment and the storage medium of diagram data | |
| CN113378812A (en) | Digital dial plate identification method based on Mask R-CNN and CRNN | |
| CN114581502A (en) | Monocular image-based three-dimensional human body model joint reconstruction method, electronic device and storage medium | |
| CN115115552A (en) | Image correction model training method, image correction device and computer equipment | |
| CN113221695B (en) | Method for training skin color recognition model, method for recognizing skin color and related device | |
| CN116701706B (en) | Data processing method, device, equipment and medium based on artificial intelligence | |
| CN114638744B (en) | Human body posture migration method and device | |
| CN116665219A (en) | Data processing method and device | |
| CN114821811B (en) | Method and device for generating person composite image, computer device and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |