CN109045664B - Diving scoring method, server and system based on deep learning - Google Patents
Diving scoring method, server and system based on deep learning Download PDFInfo
- Publication number
- CN109045664B CN109045664B CN201811030493.8A CN201811030493A CN109045664B CN 109045664 B CN109045664 B CN 109045664B CN 201811030493 A CN201811030493 A CN 201811030493A CN 109045664 B CN109045664 B CN 109045664B
- Authority
- CN
- China
- Prior art keywords
- model
- diving
- scoring
- training
- deep learning
- 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.)
- Expired - Fee Related
Links
- 230000009189 diving Effects 0.000 title claims abstract description 134
- 238000013135 deep learning Methods 0.000 title claims abstract description 42
- 238000013077 scoring method Methods 0.000 title claims abstract description 12
- 238000012549 training Methods 0.000 claims abstract description 51
- 238000005457 optimization Methods 0.000 claims abstract description 8
- 238000004458 analytical method Methods 0.000 claims description 30
- 230000033001 locomotion Effects 0.000 claims description 27
- 238000000034 method Methods 0.000 claims description 24
- 230000000875 corresponding effect Effects 0.000 claims description 19
- 230000009471 action Effects 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 3
- 238000004590 computer program Methods 0.000 description 8
- 238000012545 processing Methods 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 238000013527 convolutional neural network Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000003860 storage Methods 0.000 description 4
- 238000009432 framing Methods 0.000 description 3
- 238000013480 data collection Methods 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 238000012706 support-vector machine Methods 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 2
- HPTJABJPZMULFH-UHFFFAOYSA-N 12-[(Cyclohexylcarbamoyl)amino]dodecanoic acid Chemical compound OC(=O)CCCCCCCCCCCNC(=O)NC1CCCCC1 HPTJABJPZMULFH-UHFFFAOYSA-N 0.000 description 1
- NBGBEUITCPENLJ-UHFFFAOYSA-N Bunazosin hydrochloride Chemical compound Cl.C1CN(C(=O)CCC)CCCN1C1=NC(N)=C(C=C(OC)C(OC)=C2)C2=N1 NBGBEUITCPENLJ-UHFFFAOYSA-N 0.000 description 1
- 230000004913 activation Effects 0.000 description 1
- 230000003796 beauty Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000010224 classification analysis Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- PCHJSUWPFVWCPO-UHFFFAOYSA-N gold Chemical compound [Au] PCHJSUWPFVWCPO-UHFFFAOYSA-N 0.000 description 1
- 239000010931 gold Substances 0.000 description 1
- 229910052737 gold Inorganic materials 0.000 description 1
- 230000010365 information processing Effects 0.000 description 1
- 238000003475 lamination Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000000178 monomer Substances 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- SBNFWQZLDJGRLK-UHFFFAOYSA-N phenothrin Chemical compound CC1(C)C(C=C(C)C)C1C(=O)OCC1=CC=CC(OC=2C=CC=CC=2)=C1 SBNFWQZLDJGRLK-UHFFFAOYSA-N 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000000611 regression analysis Methods 0.000 description 1
- 239000007921 spray Substances 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
- 238000012800 visualization Methods 0.000 description 1
Classifications
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
- A63B71/00—Games or sports accessories not covered in groups A63B1/00 - A63B69/00
- A63B71/06—Indicating or scoring devices for games or players, or for other sports activities
-
- A—HUMAN NECESSITIES
- A63—SPORTS; GAMES; AMUSEMENTS
- A63B—APPARATUS FOR PHYSICAL TRAINING, GYMNASTICS, SWIMMING, CLIMBING, OR FENCING; BALL GAMES; TRAINING EQUIPMENT
- A63B24/00—Electric or electronic controls for exercising apparatus of preceding groups; Controlling or monitoring of exercises, sportive games, training or athletic performances
- A63B24/0003—Analysing the course of a movement or motion sequences during an exercise or trainings sequence, e.g. swing for golf or tennis
- A63B24/0006—Computerised comparison for qualitative assessment of motion sequences or the course of a movement
-
- 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
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Physical Education & Sports Medicine (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of diving scoring method, server and system based on deep learning.Wherein, a kind of diving scoring method based on deep learning, comprising: building diving scoring model；The diving scoring model includes C3DNet model, PoseNet model and SVR support vector regression model, and the C3DNet model and PoseNet model are connected in parallel and then are connected in series with SVR support vector regression model；The C3DNet model is realized based on the Caffe frame of Intel optimization；Utilize known diving sets of video data and corresponding diving fractional value collection, training diving scoring model；It will dive in the diving scoring model that video input is completed to training, and export diving fractional value.Which obviate the Human disturbances of diving fractional value, improve the accuracy of diving fractional value.
Description
Technical field
The invention belongs to field of information processing more particularly to a kind of diving scoring methods based on deep learning, clothes of diving
Business device and system.
Background technique
Diving refers to sporter's take-off on the instrument of certain altitude, is one terminated to enter water after completing action in the air
Item acrobatic gymnastics waterborne.Diving generally can be divided into contest diving and non-contest diving two major classes.However, for contest
Property diving often there is diving race unfairness and dispute on phenomenon, and some unjust phenomenons after match it is obvious that all can
Huge public opinion is caused to be argued.Such as in the men's 3m springboard diving match that 17 morning of 2008 Beijing time August carries out,
Chinese player Peng Bo four-wheel before match carries out final jump in the case where having a head start, and judges intentional penalty China Team and loses
Accidentally, gold medal is caused to be lost.Certain judge's real name disclose diving golds all default, oneself and judge for another example in the National Games in 10 years
Group is manipulated.
In addition, culture judge is time-consuming and expends substantial contribution.One outstanding judge generally requires tens of years and goes to train
It supports, judge needs to go ceaselessly to learn again in the meantime, and bout generally requires 5-7 judges, employs the expense of judge
With expenditure important often and in contest fund.
Race marking of diving is regular are as follows: the judge of racing dive has 7 people and 5 people's systems, then plus 1 chief referee.But the Olympic Games, generation
Boundary's championship and World Cup Competition must have 7 judge scorings.The full marks of each movement are 10 points.When scoring, leave out highest and
Minimum point, remaining score is added multiplied by difficulty point, just obtains the score of the movement.Judge can be according to the run-up of sportsman
It is (i.e. andante, treadmill), take-off, aerial and enter hydrodynamic(al) and make to evaluate score.Scoring is mainly made of difficulty point with judgers,
Difficulty point is absolutely objective, but judgers are entirely to be made of the subjective marking of judge.Although judge's marking has certain detailed rules and regulations,
It but is more still to rely on judge personal experience.And this experience can be unreliable, marking result, which can convince people, all to remain to be discussed.
In conclusion being made of at present about some in the marking of racing dive the subjective marking of judge, obtain in this way
Marking result be inaccurate, and to sportsman and unfair.Therefore, it is urgent to provide a kind of scoring methods of accurately and efficiently diving
And system.
Summary of the invention
In order to solve the deficiencies in the prior art, the first object of the present invention is to provide a kind of diving based on deep learning and beats
Divide method, deep learning is firstly introduced race marking field, can be improved the accuracy of diving appraisal result.
A kind of diving scoring method based on deep learning of the invention, comprising:
Building diving scoring model；The diving scoring model includes that C3DNet model, PoseNet model and SVR are supported
Vector regression model, the C3DNet model and PoseNet model be connected in parallel and then with SVR support vector regression model
It is connected in series；The C3DNet model is realized based on the Caffe frame of Intel optimization；
Utilize known diving sets of video data and corresponding diving fractional value collection, training diving scoring model；
It will dive in the diving scoring model that video input is completed to training, and export diving fractional value.
Further, this method further includes:
C3DNet model and PoseNet model are connected in parallel and then are connected in series with SVM classifier, it is dynamic to constitute
Perform an analysis model；
Result tally set, training action analysis model are analyzed using known diving sets of video data and corresponding actions；
It will dive in the motion analysis model that video input is completed to training, export dive comment.
Further, this method further includes:
Utilize known diving sets of video data and corresponding trunk recognition result collection, training PoseNet model；
It will dive in the PoseNet model that video input is completed to training, export trunk recognition result.
The second object of the present invention is to provide a kind of diving marking server based on deep learning.
A kind of diving marking server based on deep learning of the invention, comprising:
Scoring model of diving constructs module, is configured as: building diving scoring model；The diving scoring model includes
C3DNet model, PoseNet model and SVR support vector regression model, the C3DNet model and PoseNet model parallel connection connect
It connects and then is connected in series with SVR support vector regression model；The C3DNet model is the Caffe frame based on Intel optimization
Frame is realized；
It dives scoring model training module, is configured as: score of using known diving sets of video data and accordingly diving
Value collection, training diving scoring model；
Diving fractional value output module, is configured as: the diving scoring model that the video input that will dive is completed to training
In, output diving fractional value.
Further, the server further include:
Motion analysis model construction module, is configured as: C3DNet model and PoseNet model are connected in parallel it
Afterwards, then with SVM classifier it is connected in series, to constitute motion analysis model；
Motion analysis model training module, is configured as: being analyzed using known diving sets of video data and corresponding actions
As a result tally set, training action analysis model；
Dive comment output module, is configured as: the motion analysis mould that video input of diving is completed to training
In type, dive comment is exported.
Further, the server further include:
PoseNet model training module, is configured as: utilizing known diving sets of video data and corresponding trunk identification knot
Fruit collection, training PoseNet model；
Trunk recognition result output module, is configured as: the PoseNet model that video input of diving is completed to training
In, export trunk recognition result.
The third object of the present invention is to provide a kind of diving scoring system based on deep learning.
A kind of diving scoring system based on deep learning of the invention, including the jump described above based on deep learning
Water marking server.
Further, the diving scoring system based on deep learning, further includes client, be based on deep learning
Diving marking server be connected, for shows based on deep learning diving give a mark server output result.
Compared with prior art, the beneficial effects of the present invention are:
(1) C3DNet model of the invention is realized based on the Caffe frame of Intel optimization；In original-pack Intel
The new convolutional layer for being directly used in reading video data and 3 dimensions is compiled on Caffe, in this way without using framing and traditional volume
Lamination operates frame, can directly be handled four-dimensional array (blob) with Three dimensional convolution, improve data processing speed.
(2) present invention uses the thought of integrated study, portrays the feature of diving with two networks respectively, C3D network is portrayed
Temporal aspect, PoseNet network portray single frames human body information, most merge to obtain final score through SVR afterwards, improve last diving
The accuracy of fractional value.
(3) present invention introduces the methods of multi-task learning, in shallow-layer parameter, all type shared parameters, last
SVR on each network is respectively trained, be a kind of multi-task learning method based on parameter sharing, finally effectively improve net
Network performance.
Detailed description of the invention
The accompanying drawings constituting a part of this application is used to provide further understanding of the present application, and the application's shows
Meaning property embodiment and its explanation are not constituted an undue limitation on the present application for explaining the application.
Fig. 1 is a kind of embodiment flow chart of the diving scoring method of the invention based on deep learning.
Fig. 2 is a kind of example structure schematic diagram of the diving marking server of the invention based on deep learning.
Specific embodiment
It is noted that following detailed description is all illustrative, it is intended to provide further instruction to the application.Unless another
It indicates, all technical and scientific terms used herein has usual with the application person of an ordinary skill in the technical field
The identical meanings of understanding.
It should be noted that term used herein above is merely to describe specific embodiment, and be not intended to restricted root
According to the illustrative embodiments of the application.As used herein, unless the context clearly indicates otherwise, otherwise singular
Also it is intended to include plural form, additionally, it should be understood that, when in the present specification using term "comprising" and/or " packet
Include " when, indicate existing characteristics, step, operation, device, component and/or their combination.
Technical term is explained:
(1) deep neural network: deep neural network is the artificial neural network with multilayer neuron, in input layer and
There are multiple hidden layers between output layer.Data can be mutually transmitted between each layer of neuron, and according to the function of network
Target dynamic adjusts the weighted value of itself.
(2) C3DNet model: 3 for video feature extraction tie up convolutional network.
(3) PoseNet model: PoseNet is the vision positioning model that Cambridge University does, and can pass through a Zhang Caise
Your posture information of framing.Under a big urban environment, the posture information that it only needs to spend 5ms to obtain you,
Ratio of precision GPS is higher.And compared to GPS, it can determine your direction and can run indoors.PoseNet is to use sword
Bridge landmark data collection is trained.It is a large-scale urban environment location data collection, and having this is more than 12000 Cambridge weeks
The 6 class scene images enclosed, every image all correspond to the camera posture information of 6DOF.
(4) SVR support vector regression model: (i.e. input label is the method that SVR () is exactly SVM algorithm to do recurrence
Method when successive value).
(5) SVM:Support Vector Machine, refers to support vector machines, is a kind of common method of discrimination.
It is the learning model for having supervision in machine learning field, commonly used to carry out pattern-recognition, classification and regression analysis.
(6) Caffe:Caffe is the deep learning frame of BVLC exploitation, is based on C++ and CUDA C Plus Plus, and provide
Python and Matlab interface；The frame has convolutional neural networks CNN, Recognition with Recurrent Neural Network RNN and multilayer perceptron very much
It helps.
Deep learning was persistently had an effect in recent years, even achieves in the contest ImageNet of computer vision and is more than
The performance of mankind's average level.And deep learning is also current state of theart (best) for the processing of video
Performance.But deep learning is combined with diving marking and is had the following problems:
(1) data set count issue: diving marking problem is a very novel field, is not increased income on public network
Tape label data set, it is therefore desirable to voluntarily acquire.It is limited to time, financial resources, the sample size well marked only has more than 2000,
This is too small for deep learning, and directly training will lead to serious over-fitting, gradient can not normal backpropagation, net
Network can not be trained normally.
(2) continuously diving feature difficulty is portrayed: dive is an extremely complex continuous action, and sportsman is in high speed
Movement in, using CNN (Convolutional Neural Network, convolutional neural networks), VGG (Visual
Geometry Group, large-scale image identification depth convolutional network) even ResNet (depth residual error network) capture this
Performance is all unsatisfactory when a little features.
(3) it dives many kinds of, corresponding standards of grading are also various: different movement types has different scoring marks in diving
Standard, may also be different even if acting identical score.However data are inherently very rare, if training one to each movement
Model, will be so that data be more sparse.
Problem above will lead to diving appraisal result inaccuracy, and then occur to the inequitable phenomenon of sportsman.
To solve the above-mentioned problems, deep learning is firstly introduced race marking field by the application, be can be improved diving and is commented
Divide the accuracy of result.
As shown in Figure 1, a kind of diving scoring method based on deep learning of the present embodiment, comprising:
Step 1: building diving scoring model；The diving scoring model include C3DNet model, PoseNet model and
SVR support vector regression model, the C3DNet model and PoseNet model are connected in parallel and then return with SVR supporting vector
Model is returned to be connected in series；The C3DNet model is realized based on the Caffe frame of Intel optimization；
Step 2: utilizing known diving sets of video data and corresponding diving fractional value collection, training diving scoring model；
Step 3: in the diving scoring model that video input is completed to training that will dive, output diving fractional value.
Since PoseNet model can accurately capture the posture feature of human body, the posture of PoseNet model catcher's body
It is characterized in well making up for the information that can not be captured to C3DNet model.
For C3DNet model and PoseNet model fusion use SVR support vector regression model, C3DNet model and
Two models of PoseNet model are all directly using ground-truth training, output numerical value type score.SVR supporting vector is returned
The effect of model is returned to be to realize Ensemble, input is after the characteristic layer of two models cascades as a result, training data is exactly
True score.
Specifically, the workflow of C3DNet model are as follows:
1) video will become the array of one 4 dimension after reading, and respectively indicate channel, frame number, length and width.
2) convolution operation is carried out to array with 3d convolution, obtains the array of one 2 dimension.
3) array for 2 dimensions that step 2) obtains obtains a vector by two-way LSTM.
4) vector that step 3) obtains passes through a full articulamentum, obtains a numerical value, i.e. diving score.
The workflow of PoseNet model:
1) with the openpose tool of open source, (this tool is increased income on the net, and specific implementation is very complicated, we are only
Simply to bring application) video is handled, can obtain the coordinate of the human body key point of each frame and treated view
Frequently.
2) coordinate of each frame human body key point is sent into two-way GRU network, obtains a vector.
3) vector becomes a numerical value by a full articulamentum, is diving score.
In another embodiment, this method further include:
C3DNet model and PoseNet model are connected in parallel and then are connected in series with SVM classifier, it is dynamic to constitute
Perform an analysis model；
Result tally set, training action analysis model are analyzed using known diving sets of video data and corresponding actions；
It will dive in the motion analysis model that video input is completed to training, export dive comment.
The workflow of motion analysis model are as follows:
1) video will become the array of one 4 dimension after reading, and respectively indicate channel, frame number, length and width.
2) convolution operation is carried out to array with 3d convolution, obtains the array of one 2 dimension.
3) array for 2 dimensions that step 2) obtains obtains a vector by two-way LSTM.
4) vector that step 3) obtains passes through sigmoid activation primitive, each numerical value in vector becomes 0 or 1.
If it is 1, mean that corresponding comment is selected.
Motion analysis model is used to analyze the standard and nonstandard place of dive.
Realize that the key of motion analysis model is data, the tally set of motion analysis result is 6 common labels,
Such as: take-off opportunity is too early, and spray is excessive etc..Each video has 4 labels, so motion analysis model conversation is that a classification is asked
Topic, wherein feature extraction is carried out using with C3DNet model consistent in FocNet model, last SVR support vector regression
Model is changed to SVM classifier to realize motion analysis model.
In another embodiment, this method further include:
Utilize known diving sets of video data and corresponding trunk recognition result collection, training PoseNet model；
It will dive in the PoseNet model that video input is completed to training, export trunk recognition result.
Trunk recognition result goes out the trunk and key point of personage for identification.
On the one hand trunk identification model provides Partial Feature for scoring model, on the other hand its feature visualization result
Also can show on platform.PoseNet is that the OpenPose progress fine-tuning based on Carnegie Mellon University's open source is (micro-
Adjust), due to OpenPose had reached with high accuracy as a result, only mark a small amount of diving video can obtain it is very complete
The result of beauty.
Such as: the performance in this index is 0.5, it is also assumed that for extensive sample, score that model provides
0.5 is not above with the mathematic expectaion of the gap of true value；On the other hand, whether related coefficient is accurate for measuring ranking
Index is 0.87 in the performance of the index.Select the race video of the 2012 London Olympic Games as test set, for the Olympic Games
The judge of meeting, will be tested with judge of the same index to marking, and the Olympic Games share seven judges, to seven judges point
It does not test, it is found that the performance of this method of the invention in average absolute value error has surmounted most of monomers judge, occupy the
Two, third is occupied on related coefficient.It is considered that having reached the level of same expert.
The server as shown in Fig. 2, a kind of diving based on deep learning of the invention is given a mark, comprising:
(1) diving scoring model constructs module, is configured as: building diving scoring model；The diving scoring model
Including C3DNet model, PoseNet model and SVR support vector regression model, the C3DNet model and PoseNet model are simultaneously
Connection is connected and then is connected in series with SVR support vector regression model；The C3DNet model is optimized based on Intel
Caffe frame is realized；
(2) diving scoring model training module, is configured as: utilizing known diving sets of video data and corresponding diving point
Set of values, training diving scoring model；
(3) diving fractional value output module, is configured as: the diving marking mould that video input to the training that will dive is completed
In type, output diving fractional value.
In another embodiment, the server further include:
Motion analysis model construction module, is configured as: C3DNet model and PoseNet model are connected in parallel it
Afterwards, then with SVM classifier it is connected in series, to constitute motion analysis model；
Motion analysis model training module, is configured as: being analyzed using known diving sets of video data and corresponding actions
As a result tally set, training action analysis model；
Dive comment output module, is configured as: the motion analysis mould that video input of diving is completed to training
In type, dive comment is exported.
In another embodiment, the server further include:
PoseNet model training module, is configured as: utilizing known diving sets of video data and corresponding trunk identification knot
Fruit collection, training PoseNet model；
Trunk recognition result output module, is configured as: the PoseNet model that video input of diving is completed to training
In, export trunk recognition result.
The present invention also provides a kind of diving scoring system based on deep learning.
A kind of diving scoring system based on deep learning of the invention, including as shown in Figure 2 based on deep learning
Diving marking server.
In specific implementation, the diving scoring system based on deep learning, further includes client, be based on depth
The diving marking server of study is connected, for showing the output result of the diving marking server based on deep learning.
C3DNet model of the invention is realized based on the Caffe frame of Intel optimization；In original-pack IntelCaffe
The new convolutional layer for being directly used in reading video data and 3 dimensions of upper compiling, in this way without using framing and traditional convolutional layer
Frame is operated, directly four-dimensional array (blob) can be handled with Three dimensional convolution, improve data processing speed.
The present invention uses the thought of integrated study, portrays the feature of diving with two networks respectively, C3D network portrays timing
Feature, PoseNet network portray single frames human body information, most merge to obtain final score through SVR afterwards, improve last diving score
The accuracy of value.
Present invention introduces the methods of multi-task learning, in shallow-layer parameter, all type shared parameters, last
Each network is respectively trained on SVR, is a kind of multi-task learning method based on parameter sharing, finally effectively improves network
Performance.
It should be understood by those skilled in the art that, the embodiment of the present invention can provide as method, system or computer program
Product.Therefore, the shape of hardware embodiment, software implementation or embodiment combining software and hardware aspects can be used in the present invention
Formula.Moreover, the present invention, which can be used, can use storage in the computer that one or more wherein includes computer usable program code
The form for the computer program product implemented on medium (including but not limited to magnetic disk storage and optical memory etc.).
The present invention be referring to according to the method for the embodiment of the present invention, the process of equipment (system) and computer program product
Figure and/or block diagram describe.It should be understood that every one stream in flowchart and/or the block diagram can be realized by computer program instructions
The combination of process and/or box in journey and/or box and flowchart and/or the block diagram.It can provide these computer programs
Instruct the processor of general purpose computer, special purpose computer, Embedded Processor or other programmable data processing devices to produce
A raw machine, so that being generated by the instruction that computer or the processor of other programmable data processing devices execute for real
The device for the function of being specified in present one or more flows of the flowchart and/or one or more blocks of the block diagram.
These computer program instructions, which may also be stored in, is able to guide computer or other programmable data processing devices with spy
Determine in the computer-readable memory that mode works, so that it includes referring to that instruction stored in the computer readable memory, which generates,
Enable the manufacture of device, the command device realize in one box of one or more flows of the flowchart and/or block diagram or
The function of being specified in multiple boxes.
These computer program instructions also can be loaded onto a computer or other programmable data processing device, so that counting
Series of operation steps are executed on calculation machine or other programmable devices to generate computer implemented processing, thus in computer or
The instruction executed on other programmable devices is provided for realizing in one or more flows of the flowchart and/or block diagram one
The step of function of being specified in a box or multiple boxes.
Those of ordinary skill in the art will appreciate that realizing all or part of the process in above-described embodiment method, being can be with
Relevant hardware is instructed to complete by computer program, the program can be stored in a computer-readable storage medium
In, the program is when being executed, it may include such as the process of the embodiment of above-mentioned each method.Wherein, the storage medium can be magnetic
Dish, CD, read-only memory (Read-Only Memory, ROM) or random access memory (Random
AccessMemory, RAM) etc..
Above-mentioned, although the foregoing specific embodiments of the present invention is described with reference to the accompanying drawings, not protects model to the present invention
The limitation enclosed, those skilled in the art should understand that, based on the technical solutions of the present invention, those skilled in the art are not
Need to make the creative labor the various modifications or changes that can be made still within protection scope of the present invention.
Claims (8)
1. a kind of diving scoring method based on deep learning characterized by comprising
Building diving scoring model；The diving scoring model includes C3DNet model, PoseNet model and SVR supporting vector
Regression model, the C3DNet model and PoseNet model are connected in parallel and then connect with SVR support vector regression model
Connection；The C3DNet model is realized based on the Caffe frame of Intel optimization；
Utilize known diving sets of video data and corresponding diving fractional value collection, training diving scoring model；
It will dive in the diving scoring model that video input is completed to training, and export diving fractional value.
2. a kind of diving scoring method based on deep learning as described in claim 1, which is characterized in that this method is also wrapped
It includes:
C3DNet model and PoseNet model are connected in parallel and then are connected in series with SVM classifier, with composition movement point
Analyse model；
Result tally set, training action analysis model are analyzed using known diving sets of video data and corresponding actions；
It will dive in the motion analysis model that video input is completed to training, export dive comment.
3. a kind of diving scoring method based on deep learning as described in claim 1, which is characterized in that this method is also wrapped
It includes:
Utilize known diving sets of video data and corresponding trunk recognition result collection, training PoseNet model；
It will dive in the PoseNet model that video input is completed to training, export trunk recognition result.
The server 4. a kind of diving based on deep learning is given a mark characterized by comprising
Scoring model of diving constructs module, is configured as: building diving scoring model；The diving scoring model includes
C3DNet model, PoseNet model and SVR support vector regression model, the C3DNet model and PoseNet model parallel connection connect
It connects and then is connected in series with SVR support vector regression model；The C3DNet model is the Caffe frame based on Intel optimization
Frame is realized；
Diving scoring model training module, is configured as: known diving sets of video data and corresponding diving fractional value collection are utilized,
Training diving scoring model；
Diving fractional value output module, is configured as: defeated in the diving scoring model that video input is completed to training that will dive
It dives out fractional value.
The server 5. a kind of diving based on deep learning as claimed in claim 4 is given a mark, which is characterized in that the server is also
Include:
Motion analysis model construction module, is configured as: C3DNet model and PoseNet model are connected in parallel and then
It is connected in series with SVM classifier, to constitute motion analysis model；
Motion analysis model training module, is configured as: analyzing result using known diving sets of video data and corresponding actions
Tally set, training action analysis model；
Dive comment output module, is configured as: in the motion analysis model that video input is completed to training that will dive,
Export dive comment.
The server 6. a kind of diving based on deep learning as claimed in claim 4 is given a mark, which is characterized in that the server is also
Include:
PoseNet model training module, is configured as: utilizing known diving sets of video data and corresponding trunk recognition result
Collection, training PoseNet model；
Trunk recognition result output module, is configured as: in the PoseNet model that video input is completed to training that will dive,
Export trunk recognition result.
7. a kind of diving scoring system based on deep learning, which is characterized in that including as described in any one of claim 4-6
Diving based on deep learning give a mark server.
8. a kind of diving scoring system based on deep learning as claimed in claim 7, which is characterized in that further include client
End is connected with the diving marking server based on deep learning, for showing the diving marking server based on deep learning
Output result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811030493.8A CN109045664B (en) | 2018-09-05 | 2018-09-05 | Diving scoring method, server and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811030493.8A CN109045664B (en) | 2018-09-05 | 2018-09-05 | Diving scoring method, server and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109045664A CN109045664A (en) | 2018-12-21 |
| CN109045664B true CN109045664B (en) | 2019-10-01 |
Family
ID=64758454
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811030493.8A Expired - Fee Related CN109045664B (en) | 2018-09-05 | 2018-09-05 | Diving scoring method, server and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109045664B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110276380B (en) * | 2019-05-22 | 2021-08-17 | 杭州电子科技大学 | Real-time motion on-line guidance system based on depth model framework |
| CN110222977A (en) * | 2019-06-03 | 2019-09-10 | 张学志 | One kind movement sport methods of marking based on computer vision and device |
| CN110327611A (en) * | 2019-06-17 | 2019-10-15 | 杭州电子科技大学 | A kind of platform diving evaluation method that athletic type's coefficient is added |
| CN110314346A (en) * | 2019-07-03 | 2019-10-11 | 重庆道吧网络科技有限公司 | Intelligence fistfight sports series of skills and tricks in boxing, footmuff, system and method based on big data analysis |
| CN110354483A (en) * | 2019-07-08 | 2019-10-22 | 重庆道吧网络科技有限公司 | A kind of intelligent protector and system of grappling |
| CN118015708A (en) * | 2024-04-08 | 2024-05-10 | 华侨大学 | Diving movement quality assessment method, device and equipment based on judge score learning |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107392252A (en) * | 2017-07-26 | 2017-11-24 | 上海城诗信息科技有限公司 | Computer deep learning characteristics of image and the method for quantifying perceptibility |
| CN108288475A (en) * | 2018-02-12 | 2018-07-17 | 成都睿码科技有限责任公司 | A kind of sports video collection of choice specimens clipping method based on deep learning |
-
2018
- 2018-09-05 CN CN201811030493.8A patent/CN109045664B/en not_active Expired - Fee Related
Also Published As
| Publication number | Publication date |
|---|---|
| CN109045664A (en) | 2018-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109045664B (en) | Diving scoring method, server and system based on deep learning | |
| EP4002198A1 (en) | Posture acquisition method and device, and key point coordinate positioning model training method and device | |
| Host et al. | An overview of Human Action Recognition in sports based on Computer Vision | |
| CN105190699B (en) | Karaoke incarnation animation based on facial motion data | |
| US10803762B2 (en) | Body-motion assessment device, dance assessment device, karaoke device, and game device | |
| CN107423398A (en) | Exchange method, device, storage medium and computer equipment | |
| CN107341434A (en) | Processing method, device and the terminal device of video image | |
| CN108205684A (en) | Image disambiguation method, device, storage medium and electronic equipment | |
| CN105912991B (en) | Activity recognition based on 3D point cloud and crucial bone node | |
| CN105561535B (en) | A kind of physical fitness management system based on technology of Internet of things | |
| CN108499107A (en) | The control method of virtual role, device and storage medium in virtual reality | |
| CN109034090A (en) | A kind of emotion recognition system and method based on limb action | |
| CN110047081A (en) | Example dividing method, device, equipment and the medium of chest x-ray image | |
| Liu | Objects detection toward complicated high remote basketball sports by leveraging deep CNN architecture | |
| CN110263822A (en) | A kind of Image emotional semantic analysis method based on multi-task learning mode | |
| CN109978074A (en) | Image aesthetic feeling and emotion joint classification method and system based on depth multi-task learning | |
| Luo et al. | Player experience extraction from gameplay video | |
| Akhter | Automated posture analysis of gait event detection via a hierarchical optimization algorithm and pseudo 2D stick-model | |
| Thawonmas et al. | Artificial general intelligence in games: Where play meets design and user experience | |
| Deshpande et al. | C-STS: Conditional semantic textual similarity | |
| CN105224952B (en) | Double interbehavior recognition methods based on largest interval markov pessimistic concurrency control | |
| Han et al. | [Retracted] Martial Arts Routine Training Method Based on Artificial Intelligence and Big Data of Lactate Measurement | |
| CN115346160A (en) | Physical education teaching method, system, equipment and medium | |
| CN115475373A (en) | Motion data display method and device, storage medium and electronic device | |
| Nian et al. | [Retracted] Motion Capture Based on Intelligent Sensor in Snow and Ice Sports |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20191001 |