Intelligent training model building method and device, training method and device
Technical field
The present invention relates to a kind of intelligent training model building method and device, training method and devices, belong to people
Work intelligent image identification technology field.
Background technique
In sport training process, the normalization of movement is most important, and training is not only not achieved in nonstandard movement
Effect, but also may cause injury gained in sports, however, it is too high to ask the coach of profession to carry out guidance cost;It is traditional wearable to set
It is standby to use contact deployed with devices on the body of user, it is not only cumbersome, need user to cooperate on one's own initiative, there are certain limitations, more
It is important that cost problem cannot solve.
Although existing currently, problem above can be solved to a certain extent by knowing method for distinguishing based on human action
Other problems: one, without be directed to different figure, corresponding sport training plan is provided.Two, it is acted by action recognition judgement
Whether standardize, this method only carries out a judgement and provides whole amendment scheme, can not precise positioning to each of body
Position, the movement for not recognizing which position of human body specifically is lack of standardization, to provide correction scheme.
Summary of the invention
The first purpose of this invention is to provide a kind of intelligent training model building method, and this method can be with structure
It builds out figure identification model, specification posture action recognition model, limb action identification model and corresponding limb action and corrects model,
To be called before user movement training and in sport training process.
Second object of the present invention is to provide a kind of intelligent sports training method, and this method can be in user movement
Before training, the figure of user is identified, preliminary sport training plan is provided according to the figure of user, makes the fortune of user
Dynamic training is more scientific, and in user movement training process, is navigated to by human body attitude identification and acts inconsistent tool
Body region accurately prompts user to make correction.
Third object of the present invention is to provide a kind of intelligent training model construction device.
Fourth object of the present invention is to provide a kind of intelligent motion training device.
Of the invention the 5th is designed to provide a kind of calculating equipment.
Of the invention the 6th is designed to provide a kind of storage medium.
The first purpose of this invention can be reached by adopting the following technical scheme that:
A kind of intelligence training model building method, which comprises
Obtain the specification posture action data marked;
The specification posture action data marked by deep learning training, obtains specification posture action recognition model;Its
In, the specification posture action recognition model for identification user posture movement, judge user posture movement whether standardize;
Obtain the limb action data marked;
The limb action data marked by deep learning training, obtain limb action identification model and corresponding limbs are dynamic
Make correction model;Wherein, the limb action identification model is used to identify lack of standardization when the movement of the posture of user is lack of standardization
Body part and movement, the limb action correct model and be used to be provided corresponding according to nonstandard body part and movement
Limb action correction scheme.
Further, the method also includes:
Obtain the figure data marked;
The figure data marked by deep learning training, obtain figure identification model;Wherein, the figure identifies mould
The figure of type user for identification.
Further, it is described obtain the figure data that have marked before, further includes: obtain different building shape image and/or
Video obtains different figure data by image segmentation;Respond input the first annotation command, to different figure data into
Rower note;
Before the specification posture action data that the acquisition has marked, further includes: obtain the posture action diagram of various specifications
Picture and/or video, obtain the different specification posture action data of human body by image segmentation;Second mark life of response input
It enables, the specification posture action data different to human body is labeled;
It is described obtain the limb action data that have marked before, further includes: obtain various specifications limb action image and/
Or video, the different limb action data of human body are obtained by image segmentation;The third annotation command of input is responded, not to human body
Same limb action data are labeled.
Further, the method also includes:
It obtains according to the different sport training plans for having marked the input of figure data, is counted by deep learning training
Draw building model.
Second object of the present invention can be reached by adopting the following technical scheme that:
A kind of intelligent sports training method based on above-mentioned training model, which comprises
Obtain the video image of user movement training;
The Video Image Segmentation of user movement training is gone out to each posture motion images of user;
The specification posture action recognition model trained is called, user's posture motion images after segmentation are identified,
Judge whether the posture movement of user standardizes;
When the movement of the posture of user is lack of standardization, the limb action identification model trained and limb action is called to correct mould
Type identifies nonstandard body part and movement, reminds user and provides corresponding limb action correction scheme.
Further, before the video image for obtaining user movement training, further includes:
Obtain user images;
By carrying out recognition of face to user images, judges whether user carries out training for the first time, instructed if so, calling
Experienced figure identification model identifies the figure of user in user images, provides corresponding training meter according to the figure of user
It draws, guidance user carries out training;Wherein, judge that user is to carry out training for the first time to refer to the user not being binding movement
The registered users of drill program.
It is further, described by judging whether user carries out training for the first time to user images progress recognition of face,
It specifically includes:
Extract the face characteristic information of user images;
The face characteristic information is matched with the facial image of all registered users, if the face characteristic is believed
All it fails to match for breath and the facial image of all registered users, then judges that user carries out training for the first time;If it exists with institute
The registered users facial image for stating face characteristic information successful match, then judging user not is first progress training.
Further, the method also includes:
The nonstandard posture movement of user is stored, and is sent to user terminal.
Third object of the present invention can be reached by adopting the following technical scheme that:
A kind of intelligence training model construction device, described device include:
Specification posture action data obtains module, for obtaining the specification posture action data marked;
Specification posture action data training module, the specification posture for having been marked by deep learning training act number
According to obtaining specification posture action recognition model;Wherein, the posture of specification posture action recognition model user for identification is dynamic
Make, judges whether the posture movement of user standardizes;
Limb action data acquisition module, for obtaining the limb action data marked;
Limb action data training module, the limb action data marked by deep learning training, it is dynamic to obtain limbs
Make identification model and corresponding limb action corrects model;Wherein, the limb action identification model is used for when the posture of user is dynamic
When making lack of standardization, nonstandard body part and movement are identified, the limb action is corrected model and is used for according to nonstandard
Body part and movement provide corresponding limb action correction scheme.
Fourth object of the present invention can be reached by adopting the following technical scheme that:
A kind of intelligent motion training device based on above-mentioned training model, described device include:
Image collection module, for obtaining the video image of user movement training;
Divide module, for the Video Image Segmentation of user movement training to be gone out to each posture motion images of user;
Judgment module acts user's posture after segmentation for calling the specification posture action recognition model trained
Image is identified, judges whether the posture movement of user standardizes;
Identification module, for when the movement of the posture of user is lack of standardization, call the limb action identification model trained with
Limb action corrects model, identifies nonstandard body part and movement, reminds user and provides corresponding limb action and entangles
Direction-determining board.
5th purpose of the invention can be reached by adopting the following technical scheme that:
A kind of calculating equipment, including processor and for the memory of storage processor executable program, the processing
When device executes the program of memory storage, above-mentioned intelligent training model building method is realized, or realize above-mentioned intelligence
Sports training method can be changed.
6th purpose of the invention can be reached by adopting the following technical scheme that:
A kind of storage medium is stored with program, when described program is executed by processor, realizes above-mentioned intelligent movement instruction
Practice model building method, or realizes above-mentioned intelligent sports training method.
The present invention have compared with the existing technology it is following the utility model has the advantages that
1, the present invention can obtain figure identification model, specification posture action recognition model, limb by deep learning training
Body action recognition model and corresponding limb action correct model, so as to before user movement training and in sport training process into
Row calls.
2, the present invention can formulate different figure data corresponding sport training plan, trained by deep learning
Model is constructed to plan, plan building model can call after the figure of identification user, so as to according to the figure number of user
According to providing corresponding sport training plan.
3, the facial image of the available user of the present invention, registers the user, and binds the movement instruction of the user
Practice plan, when the user next time carry out training when, can Direct Recognition obtain the user be registered users, thus
Provide the sport training plan of binding.
4, the present invention can be identified before user movement training by figure of the figure identification model to user,
Preliminary sport training plan is provided according to the figure of user, keeps the training of user more scientific, and in user movement
In training process, it can be instructed by specification posture action recognition model and limb action identification model with precise positioning to user movement
Nonstandard position, provides correction scheme, so that the training of user is more efficient in time when practicing.
5, the present invention can store user's nonstandard posture movement in user movement training process, and with
Be sent to user terminal after the training of family, or after user movement training by the nonstandard posture of user act into
Row storage, and it is sent to user terminal, so as to the viewing of user's subsequent calls, deepens the impression, avoid make a mistake next time.
6, the present invention can be before user movement training, by carrying out recognition of face to user images, it can be determined that use
Whether family is registered users, if it is registered users, illustrate it not and be it is first carry out training, without to its figure into
Row identification, can directly give the sport training plan of binding, and the quick setting in motion of registered users is trained.
Detailed description of the invention
In order to more clearly explain the embodiment of the invention or the technical proposal in the existing technology, to embodiment or will show below
There is attached drawing needed in technical description to be briefly described, it should be apparent that, the accompanying drawings in the following description is only this
Some embodiments of invention for those of ordinary skill in the art without creative efforts, can be with
The structure shown according to these attached drawings obtains other attached drawings.
Fig. 1 is the flow chart of the intelligent training model building method of the embodiment of the present invention 1.
Fig. 2 is to carry out in the intelligent training model building method of the embodiment of the present invention 1 to different figure data
The flow chart of mark.
Fig. 3 is specification posture different to human body in the intelligent training model building method of the embodiment of the present invention 1
The flow chart that action data is labeled.
Fig. 4 is limb action different to human body in the intelligent training model building method of the embodiment of the present invention 1
The flow chart that data are labeled.
Fig. 5 is the flow chart of the intelligent sports training method of the embodiment of the present invention 2.
Fig. 6 is the flow chart of the intelligent sports training method of the embodiment of the present invention 3.
Fig. 7 be the embodiment of the present invention 3 intelligent sports training method in judge whether user carries out training for the first time
Flow chart.
Fig. 8 is the structural block diagram of the intelligent training model construction device of the embodiment of the present invention 4.
Fig. 9 is the first segmentation module and the first mark in the intelligent training model construction device of the embodiment of the present invention 4
The correlation schematic diagram of injection molding block.
Figure 10 is the second segmentation module and the second mark in the intelligent training model construction device of the embodiment of the present invention 4
The correlation schematic diagram of injection molding block.
Figure 11 is that third divides module and third mark in the intelligent training model construction device of the embodiment of the present invention 4
The correlation schematic diagram of injection molding block.
Figure 12 is the structural block diagram of the intelligent motion training device of the embodiment of the present invention 5.
Figure 13 is the structural block diagram of the intelligent motion training device of the embodiment of the present invention 6.
Figure 14 is the structural block diagram of the calculating equipment of the embodiment of the present invention 7.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiments of the present invention, instead of all the embodiments, based on the embodiments of the present invention, ordinary skill people
Member's every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Embodiment 1:
A hot research direction of the image recognition as intelligent video analysis field, by model training, identification
It is constantly broken through in technology, has pushed the development of "smart" products.
As shown in Figure 1, present embodiments providing a kind of intelligent training model building method, this method includes following
Step:
The figure data that S101, acquisition have marked.
As shown in Fig. 2, may also include that before step S101 to obtain the figure data marked
S201, the image and/or video for obtaining different building shape, obtain different figure data by image segmentation.
The image of different building shape and/or video can be obtained by acquisition in the present embodiment, such as be shot by camera
The image and/or video of different building shape can also be obtained from database lookup, such as store up different building shape in databases in advance
Image and/or video, from database search for different building shape image and/or video can be obtained.
After obtaining the image and/or video of different building shape, a large amount of different figure images are obtained by image segmentation, are made
For for trained figure data.
The first annotation command that S202, response input, is labeled different figure data.
First annotation command of the present embodiment is inputted by the way of artificial, is specifically ordered by the first mark of staff's input
It enables, responds the first annotation command of staff input, different figure data are labeled.
After being labeled to different figure data, make different figure data with corresponding label, such as high, short,
It is fat, thin, it can segment again respectively, thus the figure data marked.
S102, the figure data marked by deep learning training, obtain figure identification model.
Deep learning is a new field in machine learning research, and motivation is that foundation, simulation human brain are divided
The neural network of study is analysed, it imitates the mechanism of human brain to explain data, and the present embodiment has been marked by deep learning training
Figure data are had the figure image of label by deep learning training, obtain figure identification model, the figure identification model
It can be called before user movement training, for identification the figure of user.
The specification posture action data that S103, acquisition have marked.
As shown in figure 3, may also include that before step S103 to obtain the specification posture action data marked
S301, the posture motion images and/or video for obtaining various specifications, obtain the different rule of human body by image segmentation
Model posture action data.
Similar with step S201, the posture motion images and/or video of the various specifications in the present embodiment can be by adopting
Collection obtains, such as the posture motion images and/or video of various specifications are shot by camera, can also obtain from database lookup
It takes, such as stores up the posture motion images and/or video of various specifications in databases in advance, various rule are searched for from database
The posture motion images and/or video of model can be obtained.
After obtaining the posture motion images and/or video of various specifications, the different rule of human body are obtained by image segmentation
Model posture motion images, as trained specification posture action data.
The second annotation command that S302, response input, the specification posture action data different to human body are labeled.
Similar with step S202, second annotation command of the present embodiment is inputted by the way of artificial, specifically by the people that works
Member's the second annotation command of input responds the second annotation command of staff input, the specification posture movement different to human body
Data are labeled.
After the specification posture action data different to human body is labeled, make the specification posture action data band that human body is different
There is corresponding label, thus the specification posture action data marked.
S104, the specification posture action data marked by deep learning training, obtain specification posture action recognition mould
Type.
The specification posture action data that the present embodiment has been marked by deep learning training, that is, pass through deep learning training band
There are the specification posture motion images of label, obtain specification posture action recognition model, which can be with
It is called in user movement training process, for identification the posture movement of user, judges whether the posture movement of user advises
Model.
The limb action data that S105, acquisition have marked.
As shown in figure 4, may also include that before step S105 to obtain the limb action data marked
S401, the limb action image and/or video for obtaining various specifications, obtain the different limb of human body by image segmentation
Body action data.
Similar with step S201, S301, the limb action image and/or video of the various specifications in the present embodiment can lead to
It crosses acquisition to obtain, such as shoots the limb action image and/or video of various specifications by camera, can also be looked into from database
Acquisition is looked for, such as stores up the limb action image and/or video of various specifications in databases in advance, is searched for from database each
The limb action image and/or video of kind specification can be obtained.
After obtaining the limb action image and/or video of various specifications, the different limb of human body is obtained by image segmentation
Body motion images, as trained limb action data.
It is appreciated that in order to reduce workload, can from the step S1031 posture motion images obtained and/or video,
Image and/or video including each position skeleton point of human body are further obtained, to obtain the limb action figure of various specifications
Picture and/or video, then obtain the different limb action data of human body by image segmentation again;It can also be directly from step
In the specification posture action data that S1031 is obtained by image segmentation, the different limb action data of human body are further obtained, it must
Image segmentation can be carried out when wanting, again to obtain satisfactory limb action data.
The third annotation command that S402, response input, the limb action data different to human body are labeled.
Similar with step S202, S302, the third annotation command of the present embodiment is inputted by the way of artificial, specifically by work
Make personnel and input third annotation command, responds the third annotation command of staff input, the limb action different to human body
Data are labeled, such as 15 skeleton point positions of mark human body, are illustrated with the skeleton point on the human body left side, respectively
Are as follows: head, two shoulder centers, left hand large-arm joint point, left hand forearm artis, left hand, waist center, left thigh artis, knee
It covers, left foot, the skeleton point on the right of human body is similarly.
After the limb action data different to human body are labeled, the limb action data for keeping human body different are with corresponding
Label, thus the limb action data marked.
S106, by the limb action data that have marked of deep learning training, obtain limb action identification model and corresponding
Limb action corrects model.
The limb action data that the present embodiment has been marked by deep learning training, i.e., by deep learning training with mark
The limb action image of label, obtains limb action identification model and corresponding limb action corrects model, which identifies mould
Type can be called in user movement training process, for identifying lack of standardization when the movement of the posture of user is lack of standardization
Body part and movement, limb action correct model be used for according to nonstandard body part and movement, provide corresponding limb
Body acts correction scheme.
Above-mentioned steps S101~S106 is off-line phase, can be applied to various different occasions, such as family's training,
Campus training, gymnasium training etc..
When being applied to family's training, the intelligence for being previously-completed above-mentioned steps S101~S106 can be placed at home
Energy electronic equipment, then kinsfolk carries out training using the intelligent electronic device;It is also possible to staff and passes through clothes
Business device is previously-completed above-mentioned steps S101~S106, and kinsfolk passes through the download services such as desktop computer, notebook computer
The intelligent training model of device, then carries out training.
When be applied to campus training when, can be placed in the sports ground and Stadium in campus be previously-completed it is above-mentioned
The intelligent electronic device of step S101~S106, then academics and students carry out training using the intelligent electronic device;?
Self-service terminal machine can be placed in the sports ground and Stadium in campus, staff is previously-completed above-mentioned steps by server
S101~S106, and self-service terminal machine is sent by intelligent training model, then academics and students use the self-service end
Terminal carries out training;It can also be that staff completes above-mentioned steps S101~S106 by self-service terminal machine, then always
Teacher and student use self-service terminal machine progress training.
When being applied to gymnasium training, every fitness equipment has corresponding human-computer interaction terminal, and each man-machine
Interactive terminal is managed collectively by server, and staff is previously-completed above-mentioned steps S101~S106 by server, and by intelligence
Training model can be changed and be sent to each human-computer interaction terminal, then body-building personnel carry out movement instruction using human-computer interaction terminal
Practice;Be also possible to staff by a wherein human-computer interaction terminal for every kind of fitness equipment complete above-mentioned steps S101~
Intelligent training model is shared with other human-computer interaction terminals of this kind of fitness equipment by S106, and then body-building personnel make
Training is carried out with the human-computer interaction terminal.
Further, the intelligent training model building method of the present embodiment may also include that
It obtains according to the different sport training plans for having marked the input of figure data, is counted by deep learning training
Draw building model;The step can execute after step slol, such as can hold after step slol, before step S102
Row is also possible to execute after the either step in step S102~S106, can also be with any in step S102~S106
Step is performed simultaneously, and specifically, staff inputs corresponding sport training plan, to not according to the figure data marked
Same figure data formulate corresponding sport training plan, obtain plan building model, the plan structure by deep learning training
Established model can call after the figure of identification user, for the figure data according to user, provide corresponding training meter
It draws.
Embodiment 2:
As shown in figure 5, present embodiments providing a kind of intelligent sports training method, fortune of this method based on embodiment 1
Dynamic training pattern is realized comprising following steps:
S501, user images are obtained.
The present embodiment obtains user images in such a way that camera is shot.
The figure identification model that S502, calling have been trained, identifies the figure of user in user images, according to the figure of user
Corresponding sport training plan is provided, guidance user carries out training.
The figure that the step S102 that figure identification model is above-described embodiment 1 training of having trained in the present embodiment obtains is known
Other model identifies that the figure of user in user images can call above-mentioned reality according to the figure of user by figure identification model
Apply example 1 plan building model, provide corresponding sport training plan, guided by way of voice and/or display user into
Row training.
It is appreciated that phase can be provided in other way if obtaining plan building model without training in advance
The sport training plan answered, such as corresponding sport training plan is bound to different figures in figure identification model in advance,
After identifying the figure of user, corresponding sport training plan can be directly given;It is instructed it is also understood that being moved in gymnasium
When practicing, camera can be set in the foregrounding of gymnasium, which is connected with foreground computer, when foreground computer is known
After the figure of other user, staff can suitably will be moved according to the figure of user by modes such as wechat, QQ, short messages
Drill program is sent to user terminal, which can be the portable mobile terminal of user, such as smart phone, palm
Any one in computer, tablet computer.
S503, user carry out training according to sport training plan, obtain the video image of user movement training.
The present embodiment obtains the view of user movement training in user movement training process in such a way that camera is shot
Frequency image.
S504, each posture motion images that the Video Image Segmentation of user movement training is gone out to user.
The specification posture action recognition model that S505, calling have been trained, carries out user's posture motion images after segmentation
Identification, judges whether the posture movement of user standardizes.
The step S104 that specification posture action recognition model is above-described embodiment 1 training of having trained in the present embodiment obtains
Specification posture action recognition model, by specification posture action recognition model to after segmentation user's posture motion images carry out
Identification, judges whether the posture movement of user standardizes, if so, user can continue movement instruction without any prompt
Practice, if it is not, entering step S506.
The limb action identification model and limb action that S506, calling have been trained correct model, identify nonstandard limb
Body region and movement remind user and provide corresponding limb action correction scheme.
Limb action identification model and limb action has been trained to correct the step that model is above-described embodiment 1 in the present embodiment
The limb action identification model and limb action that rapid S106 training obtains correct model, are identified by limb action identification model
Nonstandard body part and movement, such as arm is lifted not high enough, shoulder expansion is not in place etc., with voice and/or display
Mode reminds the nonstandard body part of user and movement, and provides corresponding limb action by limb action correction model and entangle
Direction-determining board, it is proposed that user first suspends training, enters step S507.
S507, user carry out limb action correction according to limb action correction scheme, until user corrects in limb action
When rear posture action norm, just continue subsequent training.
Above-mentioned steps S501~S507 is on-line stage, can be applied to various different occasions, such as family's training,
Campus training, gymnasium training etc..
When being applied to family's training, it can place at home and be previously-completed intelligent training model construction
Intelligent electronic device, then kinsfolk using the intelligent electronic device complete above-mentioned steps S501~S507;It can also be with work
Make personnel and intelligent training model construction is previously-completed by server, kinsfolk passes through desktop computer, notebook
Then the intelligent training model of the download servers such as computer completes above-mentioned steps S501~S507.
When being applied to campus training, it can be placed in the sports ground and Stadium in campus and be previously-completed intelligence
Change the intelligent electronic device of training model construction, then academics and students complete above-mentioned steps using the intelligent electronic device
S501~S507;Self-service terminal machine can also be placed in the sports ground and Stadium in campus, staff is pre- by server
First complete intelligent training model construction, send self-service terminal machine for intelligent training model, then teacher and
Student completes above-mentioned steps S501~S507 using the self-service terminal machine;It can also be that staff is preparatory by self-service terminal machine
Complete intelligent training model construction, then academics and students using the self-service terminal machine complete above-mentioned steps S501~
S507。
When being applied to gymnasium training, every fitness equipment has corresponding human-computer interaction terminal, and each man-machine
Interactive terminal is managed collectively by server, and staff is previously-completed intelligent training model construction by server, and
Each human-computer interaction terminal is sent by intelligent training model, then body-building personnel are completed using human-computer interaction terminal
State step S501~S507;It is also possible to staff and intelligence is completed by a wherein human-computer interaction terminal for every kind of fitness equipment
Training model construction can be changed, other human-computer interactions that intelligent training model is shared with this kind of fitness equipment are whole
End, then body-building personnel complete above-mentioned steps S501~S507 using the human-computer interaction terminal.
Further, the intelligent sports training method of the present embodiment may also include that the nonstandard posture movement of user
It is stored, and is sent to user terminal;Specifically, the storing the nonstandard posture movement of user of this step can be with step
Rapid S506 is performed simultaneously, and when the movement of the posture of user is lack of standardization, the nonstandard posture movement of user is stored in database, and
User's end is sent by the movement of nonstandard posture by modes such as wechat, QQ, short message, mails after user movement training
End;Being stored user's nonstandard posture movement of this step can also execute after user movement training, when with
After the training of family, the nonstandard posture movement of user is stored in database, and pass through wechat, QQ, short message, mail etc.
The movement of nonstandard posture is sent user terminal by mode, so as to the viewing of user's subsequent calls, deepens the impression, avoids next criminal
Mistake, as described in step S502, the user terminal can be the portable mobile terminal of user, as smart phone, palm PC,
Any one in tablet computer.
Embodiment 3:
As shown in fig. 6, present embodiments providing a kind of intelligent sports training method, fortune of this method based on embodiment 1
Dynamic training pattern is realized comprising following steps:
S601, user images are obtained.
S602, by user images carry out recognition of face, judge whether user carries out training for the first time.
The present embodiment can obtain the facial image of the user when user carries out training for the first time, to the user into
Row registration, and the sport training plan of the user is bound, specifically, when user carries out training for the first time, pass through camera
The mode of acquisition obtains the facial image of the user, registers to the user, and is intended to move accordingly in building model
Drill program is bound, therefore the facial image of a large amount of registered users is stored in database, when the user is next time
Carry out training when, can Direct Recognition obtain the user be registered users, to provide the sport training plan of binding.
Step S602 is as shown in fig. 7, specifically include:
S6021, the face characteristic information for extracting user images.
S6022, the face characteristic information of user images is matched with the facial image of all registered users, if with
All it fails to match for the facial image of the face characteristic information of family image and all registered users, then illustrates that the user is not to have infused
Volume user, judges that user carries out training for the first time, enters step S603;If it exists with the face characteristic information of user images
With successful registered users facial image, then illustrate that the user is registered users, judging user not is to move for the first time
Training directly gives the sport training plan of user binding, enters step S604;Wherein, it fails to match refers to user images
Face characteristic information and the matching similarities of facial image of registered users be less than preset threshold, successful match refers to user
The matching similarity of the facial image of the face characteristic information and registered users of image is greater than or equal to preset threshold.
The figure identification model that S603, calling have been trained, identifies the figure of user in user images, according to the figure of user
Corresponding sport training plan is provided, guidance user carries out training, enters step S604.
S604, user carry out training according to sport training plan, obtain the video image of user movement training.
S605, each posture motion images that the Video Image Segmentation of user movement training is gone out to user.
The specification posture action recognition model that S606, calling have been trained, carries out user's posture motion images after segmentation
Identification, judges whether the posture movement of user standardizes.
The limb action identification model and limb action that S607, calling have been trained correct model, identify nonstandard limb
Body region and movement remind user and provide corresponding limb action correction scheme.
S608, user carry out limb action correction according to limb action correction scheme, until user corrects in limb action
When rear posture action norm, just continue subsequent training.
Above-mentioned steps S601, S603, S604, S605, S606, S607 and S608 respectively with above-described embodiment 2 the step of
S501, S502, S503, S504, S506, S506 and S507, this is no longer going to repeat them.
Above-mentioned steps S601~S608 is on-line stage, can be applied to various different occasions, such as family's training,
Campus training, gymnasium training etc., for details, reference can be made to above-described embodiments 2, and this is no longer going to repeat them.
With above-described embodiment 2, the intelligent sports training method of the present embodiment may also include that the nonstandard posture of user
Movement is stored, and is sent to user terminal.
It will be understood by those skilled in the art that realizing that all or part of the steps in the method for above-described embodiment 1~3 can be with
Relevant hardware is instructed to complete by program, corresponding program can store in computer readable storage medium.
It should be noted that although describing the method operation of above-described embodiment 1~3 in the accompanying drawings with particular order, this
These operations must be executed in this particular order by not requiring that or implying, or is had to carry out and operated just shown in whole
It is able to achieve desired result.On the contrary, the step of describing can change and execute sequence.Additionally or alternatively, it is convenient to omit certain
Multiple steps are merged into a step and executed, and/or a step is decomposed into execution of multiple steps by step.
Embodiment 4:
As shown in figure 8, present embodiments providing a kind of intelligent training model construction device, which includes figure
Data acquisition module 801, the first training module 802, the second data acquisition module 803, the second training module 804, third data
It obtains module 805 and third training module 806, the concrete function of modules is as follows:
The figure data acquisition module 801, for obtaining the figure data marked;As shown in figure 9, in figure data
Before obtaining module 801, it may also include that
First segmentation module 901 obtains difference by image segmentation for obtaining the image and/or video of different building shape
Figure data;
First labeling module 902 is labeled different figure data for responding the first annotation command of input.
The figure data training module 802, the figure data for having been marked by deep learning training, obtains figure
Identification model;Wherein, the figure of figure identification model user for identification.
The specification posture action data obtains module 803, for obtaining the specification posture action data marked;Such as figure
Shown in 10, before specification posture action data obtains module 803, it may also include that
Second segmentation module 1001 passes through image segmentation for obtaining the posture motion images and/or video of various specifications
Obtain the different specification posture action data of human body.
Second labeling module 1002, for responding the second annotation command of input, the specification posture movement different to human body
Data are labeled.
The specification posture action data training module 804, the specification posture for having been marked by deep learning training
Action data obtains specification posture action recognition model;Wherein, specification posture action recognition model user for identification
Posture movement, judges whether the posture movement of user standardizes.
The limb action data acquisition module 805, for obtaining the limb action data marked;As shown in figure 11,
Before limb action data acquisition module 805, it may also include that
Third divides module 1101 and passes through image segmentation for obtaining the limb action image and/or video of various specifications
Obtain the different limb action data of human body.
Third labeling module 1102, for responding the third annotation command of input, the limb action data different to human body
It is labeled.
The limb action data training module 806, the limb action data for having been marked by deep learning training,
It obtains limb action identification model and corresponding limb action corrects model;Wherein, the limb action identification model is used to work as and use
When the posture movement at family is lack of standardization, nonstandard body part and movement are identified, the limb action corrects model and is used for root
According to nonstandard body part and movement, corresponding limb action correction scheme is provided.
The intelligent training model construction device of the present embodiment may also include that
Plan training module, for obtaining according to the different sport training plans for having marked the input of figure data, passes through
Deep learning training obtains plan building model.
The specific implementation of modules may refer to above-described embodiment 1 in the present embodiment, and this is no longer going to repeat them.
Embodiment 5:
As shown in figure 12, a kind of intelligent motion training device, fortune of the device based on embodiment 4 are present embodiments provided
Dynamic training pattern is realized comprising the first image collection module 1201, the first identification module 1202, the second image collection module
1203, it is as follows to divide module 1204, judgment module 1205 and the second identification module 1206, the concrete function of modules:
The first image obtains module 1201, for obtaining user images.
First identification module 1202 identifies user in user images for calling the figure identification model trained
Figure, corresponding sport training plan is provided according to the figure of user, guidance user carries out training.
Second image collection module 1203, for obtaining the video image of user movement training.
The segmentation module 1204, each posture for the Video Image Segmentation of user movement training to be gone out user act
Image.
The judgment module 1205, for calling the specification posture action recognition model trained, to the user after segmentation
Posture motion images are identified, judge whether the posture movement of user standardizes.
Second identification module 1206, for calling the limbs trained dynamic when the movement of the posture of user is lack of standardization
Make identification model and limb action corrects model, identify nonstandard body part and movement, remind user and provides corresponding
Limb action correction scheme.
The intelligent motion training device of the present embodiment, may also include that
Sending module for storing the nonstandard posture movement of user, and is sent to user terminal.
The specific implementation of modules may refer to above-described embodiment 2 in the present embodiment, and this is no longer going to repeat them.
Embodiment 6:
As shown in figure 13, a kind of intelligent motion training device, fortune of the device based on embodiment 4 are present embodiments provided
Dynamic training pattern is realized comprising image collection module 1301, first judgment module 1302, the second image collection module 1303,
It is as follows to divide module 1304, the second judgment module 1305 and identification module 1306, the concrete function of modules:
The first image obtains module 1301, for obtaining user images.
The first judgment module 1302, for judging whether user is first by carrying out recognition of face to user images
Training is carried out, if so, calling the figure identification model trained, the figure of user in user images is identified, according to user
Figure provide corresponding sport training plan, guidance user carries out training;Wherein, judge that user is to move for the first time
It is the registered users for binding sport training plan that training, which refers to the user not,.
Second image collection module 1303, for obtaining the video image of user movement training.
The segmentation module 1304, each posture for the Video Image Segmentation of user movement training to be gone out user act
Image.
Second judgment module 1305, for calling the specification posture action recognition model trained, after segmentation
User's posture motion images identify, judge whether the posture movement of user standardizes.
The identification module 1306, for calling the limb action trained to know when the movement of the posture of user is lack of standardization
Other model and limb action correct model, identify nonstandard body part and movement, remind user and provide corresponding limb
Body acts correction scheme.
The intelligent motion training device of the present embodiment, may also include that
Sending module for storing the nonstandard posture movement of user, and is sent to user terminal.
The specific implementation of modules may refer to above-described embodiment 3 in the present embodiment, and this is no longer going to repeat them.
It should be noted that the device that above-described embodiment 4~6 provides only is illustrated with the division of above-mentioned each functional module
Illustrate, in practical applications, can according to need and be completed by different functional modules above-mentioned function distribution, i.e., by internal junction
Structure is divided into different functional modules, to complete all or part of the functions described above.
It is appreciated that term " first ", " second " used in the device of the present embodiment etc. can be used for describing various modules,
But these modules should not be limited by these terms.These terms are only used to distinguish first module and another module.Citing comes
It says, without departing from the scope of the invention, first judgment module can be known as to the second judgment module, and similarly,
Second judgment module can be known as to first judgment module, first judgment module and the second judgment module both judgment module,
It but is not same judgment module.
Embodiment 7:
As shown in figure 14, a kind of calculating equipment is present embodiments provided, including the processing connected by system bus 1401
Device 1402, memory, input unit 1403, display 1404 and network interface 1405.Wherein, processor 1402 is based on providing
It calculates and control ability, memory includes non-volatile memory medium 1406 and built-in storage 1407, the non-volatile memory medium
1406 are stored with operating system, computer program and database, which is in non-volatile memory medium 1406
Operating system and computer program operation provide environment, computer program by processor 1402 execute when, realize above-mentioned reality
The intelligent training model building method of example 1 is applied, as follows:
Obtain the figure data marked;The figure data marked by deep learning training obtain figure identification mould
Type;Obtain the specification posture action data marked;The specification posture action data marked by deep learning training, obtains
Specification posture action recognition model;Obtain the limb action data marked;The limbs marked by deep learning training are dynamic
Make data, obtains limb action identification model and corresponding limb action corrects model.
Calculating equipment in the present embodiment can be server, self-service terminal machine, human-computer interaction terminal etc., can also be special
Door is used for the new intelligent electronic device of training.
Embodiment 8:
A kind of calculating equipment is present embodiments provided, is formed with embodiment 7, when computer program is executed by processor,
Realize the intelligent sports training method of above-described embodiment 2, as follows:
Obtain user images;Call the figure identification model trained, the figure of user in identification user images, according to
The figure at family provides corresponding sport training plan, and guidance user carries out training;Obtain the video figure of user movement training
Picture;The Video Image Segmentation of user movement training is gone out to each posture motion images of user;Call the specification posture trained
Action recognition model identifies user's posture motion images after segmentation, judges whether the posture movement of user standardizes;When
When the posture movement of user is lack of standardization, calls the limb action identification model trained and limb action to correct model, identify
Nonstandard body part and movement remind user and provide corresponding limb action correction scheme.
Embodiment 9:
A kind of calculating equipment is present embodiments provided, is formed with embodiment 7, when computer program is executed by processor,
Realize the intelligent sports training method of above-described embodiment 2, as follows:
Obtain user images;By judging whether user carries out training for the first time to user images progress recognition of face,
If so, calling the figure identification model trained, identifies the figure of user in user images, provided accordingly according to the figure of user
Sport training plan, guidance user carry out training;Obtain the video image of user movement training;By user movement training
Video Image Segmentation go out each posture motion images of user;The specification posture action recognition model trained is called, to dividing
User's posture motion images after cutting identify, judge whether the posture movement of user standardizes;When the posture of user acts not
When specification, calls the limb action identification model trained and limb action to correct model, identify nonstandard body part
And movement, it reminds user and provides corresponding limb action correction scheme.
Calculating equipment in above-described embodiment 8~9 can be computer, self-service terminal machine, human-computer interaction terminal etc., may be used also
To be used exclusively for the new intelligent electronic device of training.
Embodiment 10:
A kind of storage medium is present embodiments provided, which is computer readable storage medium, is stored with meter
Calculation machine program when computer program is executed by processor, realizes the intelligent training model construction side of above-described embodiment 1
Method is as follows:
Obtain the figure data marked;The figure data marked by deep learning training obtain figure identification mould
Type;Obtain the specification posture action data marked;The specification posture action data marked by deep learning training, obtains
Specification posture action recognition model;Obtain the limb action data marked;The limbs marked by deep learning training are dynamic
Make data, obtains limb action identification model and corresponding limb action corrects model.
Embodiment 11:
A kind of storage medium is present embodiments provided, which is computer readable storage medium, is stored with meter
Calculation machine program when computer program is executed by processor, realizes the intelligent sports training method of above-described embodiment 1, as follows:
Obtain user images;Call the figure identification model trained, the figure of user in identification user images, according to
The figure at family provides corresponding sport training plan, and guidance user carries out training;Obtain the video figure of user movement training
Picture;The Video Image Segmentation of user movement training is gone out to each posture motion images of user;Call the specification posture trained
Action recognition model identifies user's posture motion images after segmentation, judges whether the posture movement of user standardizes;When
When the posture movement of user is lack of standardization, calls the limb action identification model trained and limb action to correct model, identify
Nonstandard body part and movement remind user and provide corresponding limb action correction scheme.
Embodiment 12:
A kind of storage medium is present embodiments provided, which is computer readable storage medium, is stored with meter
Calculation machine program when computer program is executed by processor, realizes the intelligent sports training method of above-described embodiment 1, as follows:
Obtain user images;By judging whether user carries out training for the first time to user images progress recognition of face,
If so, calling the figure identification model trained, identifies the figure of user in user images, provided accordingly according to the figure of user
Sport training plan, guidance user carry out training;Obtain the video image of user movement training;By user movement training
Video Image Segmentation go out each posture motion images of user;The specification posture action recognition model trained is called, to dividing
User's posture motion images after cutting identify, judge whether the posture movement of user standardizes;When the posture of user acts not
When specification, calls the limb action identification model trained and limb action to correct model, identify nonstandard body part
And movement, it reminds user and provides corresponding limb action correction scheme.
Storage medium in above-described embodiment 10~12 can be disk, CD, computer storage, read-only memory
(ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), USB flash disk, mobile hard disk
Equal media.
In conclusion the present invention can obtain figure identification model, specification posture action recognition by deep learning training
Model, limb action identification model and corresponding limb action correct model, so as to before user movement training and training
It is called in the process;It can be identified by figure of the figure identification model to user, root before user movement training
Preliminary sport training plan is provided according to the figure of user, keeps the training of user more scientific, and is instructed in user movement
It, can be with precise positioning to user movement training by specification posture action recognition model and limb action identification model during white silk
When nonstandard position, correction scheme is provided in time, so that the training of user is more efficient.
The above, only the invention patent preferred embodiment, but the scope of protection of the patent of the present invention is not limited to
This, anyone skilled in the art is in the range disclosed in the invention patent, according to the present invention the skill of patent
Art scheme and its inventive concept are subject to equivalent substitution or change, belong to the scope of protection of the patent of the present invention.