CN109948459A - A kind of football movement appraisal procedure and system based on deep learning - Google Patents
A kind of football movement appraisal procedure and system based on deep learning Download PDFInfo
- Publication number
- CN109948459A CN109948459A CN201910143782.7A CN201910143782A CN109948459A CN 109948459 A CN109948459 A CN 109948459A CN 201910143782 A CN201910143782 A CN 201910143782A CN 109948459 A CN109948459 A CN 109948459A
- Authority
- CN
- China
- Prior art keywords
- football
- video
- deep learning
- skeleton
- sport video
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of, and the football based on deep learning acts appraisal procedure, comprising the following steps: S1: formulating the standard form of the Category criteria movement of various football movements；S2: camera acquisition football training, the sport video of match are utilized；S3: the video data in processing sport video obtains football action classification；S4: the standard operation of the football action classification in the football action classification matching criteria template obtained with S3, and export the difference with standard operation.The present invention proposes that a kind of human body attitude estimation model reduces the human body attitude estimation model redundancy based on deep learning, accelerate the arithmetic speed that human body attitude estimation model extracts the skeleton key point of video frame, reduce operation time, the skeleton data with space time information is constructed using skeleton point graph structure, there is characterization well for the movement of limb action local in football, discrimination improves.
Description
Technical field
The present invention relates to technical field of computer vision, act more particularly, to a kind of football based on deep learning
Appraisal procedure and system.
Background technique
Football is referred to as " the first in the world movement ", its main feature is that the number of participant is more, place is big, technology complexity,
Fixture is long, is the stronger communality sports events of an antagonism.With the development of retransmission technique and big data, increasingly
More football professional persons can by analysis compete video, statistics sportsman run distance, technical characterstic, run be accustomed to etc. features,
Statistical law is obtained by analyzing feature again, to formulate specific aim staffing, game plan.With wearable device
Develop, the data such as sportsman's running speed, thermal map of running can be obtained in football match, if but want to count and analyze player motion,
Existing method, which often passes through, manually watches video, therefore how quickly to analyze and understand the movement of sportsman by computer is one
Must very have with significant project, be using computer carry out human body attitude estimation with human action identification important application.
Football movement assessment common at present is to carry out artificial viewing video by professional person and assess, with deep learning
Development, the action recognition based on deep learning has become a very popular research side in computer vision field
Method.
A kind of human motion recognition method is proposed in Fig. 1.This method samples the video to be detected of input, obtains
The sequence of subsampled images of video to be detected, then using skeleton key point (human body attitude estimation) detection model pair trained
Sampled image frames carry out skeleton critical point detection, obtain the skeleton of each sampled image frames in sampled images frame sequence
Key point position thermal map (probability characteristics for characterizing default skeleton key point position), by the people of sampled images frame sequence
The classification of motion model that the input of body bone key point position thermal map has been trained is classified, and the corresponding human body of video to be identified is obtained
Movement.
Usually human action identifying system is mounted in PC machine in existing scheme, and football match, training are mostly outdoors
Scene carries out, and football movement assessment is also required to carry out outdoors, the use under scene outdoors of the equipment limit of existing scheme；
Skeleton critical point detection is carried out to each frame image using the skeleton critical point detection model trained, and it is existing
Skeleton critical point detection model all compares redundancy, in the detection process can it is too complicated because of model, required computing resource is big
And cause to take a long time every time, and since the scene restriction of football movement assessment makes system need to be deployed in embedded device
On, embedded device hardware resource is less with respect to the end PC, so that being difficult in actual use reaches requirement of real-time；To be checked
It surveys video to be sampled, video data is sampled into video requency frame data, then the detection of skeleton point is carried out to each frame image, and
Actually one movement is a continuous process, having time continuity between frame and frame, and the skeleton between different frame closes
Key point has space successional, and video is isolated framing processing by existing scheme, can only evaluate the movement at a certain moment, and football is dynamic
It judges and needs to evaluate an action process, above-mentioned solution does not consider space-time expending.
Summary of the invention
The present invention provides a kind of football movement appraisal procedure and system based on deep learning.
Primary and foremost purpose of the invention is to provide a kind of football movement appraisal procedure based on deep learning, solves existing model
The problem of identifying that time-consuming caused by too complicated, required computing resource is big also solves not considering that space-time connects in existing scheme
The problem of continuous property.
The further object of the present invention be to provide it is a kind of based on deep learning football movement assessment system, reduce equipment at
Sheet and operation difficulty improve portability.
In order to solve the above technical problems, technical scheme is as follows:
A kind of football movement appraisal procedure based on deep learning, comprising the following steps:
S1: the standard form of the Category criteria movement of various football movements is formulated；
S2: camera acquisition football training, the sport video of match are utilized；
S3: the video data in processing sport video obtains football action classification；
S4: the standard operation of the football action classification in the football action classification matching criteria template obtained with S3, and it is defeated
Out with the difference of standard operation.
Preferably, the standard form of various football action classification standard operations is formulated in step S1, comprising the following steps:
S1.1: it collects the open video of the standard operation of various football action classifications and carries out manual sort；
S1.2: skeleton key point is sought to the standard operation of each classification using human body attitude estimation model, and right
The skeleton key point of the standard operation of each classification is averaging, and the standard operation of each classification obtains a skeleton point sequence
Column, all skeleton point Sequence composition standard forms.
It is often used movement by observing football match, selectes 12 kinds of common foots such as chesting, foot inside trapping, heading
Ball standard operation collects the open video of such movement and carries out manual sort, reuse and trained as observation standard operation
The human body attitude estimation model of parameter carries out it to seek skeleton key point, and the bone key point of every a kind of movement is asked flat
, a bone point sequence can be obtained in every a kind of movement, which is football standard form.
Preferably, human body attitude estimates model, specific as follows:
After the video frame of sport video after normalization is input to VGG16 or ResNet convolutional neural networks, had
There is the characteristic pattern of the high dimensional feature of video frame, while being also high dimensional feature matrix；Characteristic pattern is again through LSC (Large separable
Convolution) area Proposal regions is obtained through RPN (Region Proposal Network) network again after convolution
Domain, using PSROI pooling (Position Sensitiveregion of interestPooling) to Proposal
Regions is input to R-CNN network again and carries out classification and position recurrence after being detected, R-CNN network only has one layer of full connection
Layer, finally obtains bone point sequence.
Extraction feature individually connected entirely twice to each ROI in existing method, the port number of each ROI with
Characteristic pattern is consistent, and full connection carries out inner product operation to all channels, and computationally intensive and weight cannot be shared, and can occupy a large amount of calculating
Resource and time.LSC (Large separable convolution) convolution, obtained ROI (region are set after characteristic pattern
Of interest) port number can port number in existing scheme it is few, use PSROI pooling (Position
Sensitive ROI Pooling, this word of the candidate region of position sensing) RoIAlign is replaced, in Proposal regions
It carries out manually introducing location information when pond, to be effectively improved deeper neural network to the sensitivity of object space, warp
The improvement for crossing this two step only has one layer of full connection in last R-CNN sub-network, and the port number of ROI is reduced, and is reduced complete
Inner product operation is connected, model calculation speed is accelerated.
Preferably, the video data in sport video is handled in step S3, obtains football action classification, including following step
It is rapid:
S3.1: sport video length normalization method is enabled；
S3.2: a bone point sequence is all obtained in each frame of sport video using human body attitude estimation model；
S3.3: being connected with each other in the skeleton point of each frame of sport video according to human physiological structure, same between different frame
The skeleton point of position is connected with each other, and is constructed skeleton point space-time diagram G=(V, E), wherein V is number of nodes, that is, bone points, indicates bone
The space structure of bone point, E are side, that is, feature vector in graph structure, indicate the connection between different frame, represent each specific section
The track of point over time；
S3.4: feature extraction is carried out to skeleton point space-time diagram using figure convolutional neural networks, obtains the football action classification.
Since the football training video acquired every time with camera is accustomed to by the technology of movement executor, execution movement speed
The influence of the factors such as speed, the length of the video acquired every time all have very big randomness, and therefore, it is necessary to each input
The video of sample is pre-processed, i.e. length normalization method, so that human body attitude estimates model extraction skeleton key point.Human body
It is formed by connecting by four limbs, trunk and first-class multiple positions by joint, in football technique movement, many movements are all by human body
What limbs were completed, if can determine that the relative position of partes corporis humani position, can be laid well for football training action recognition and guidance
Basis, common football movement is mostly limbs activities, and skeleton key has good solution to the space structure of human body
The property released, a movement is one section track of the limbs in time-domain, it is possible to carry out table to movement using bone point sequence
Sign.Skeleton point space-time diagram is a kind of non-Euclid's data, and the convolution algorithm in common convolutional neural networks not can be carried out fortune
It calculates, feature extraction is carried out using figure neural network.
Preferably, sport video length normalization method is enabled in step S3.1, comprising the following steps:
S3.1.1: enabling video desired length is n, judges sport video length for N, if N > n, sport video is carried out etc.
Interval sampling enables sport video length be equal to n；If N < n, interpolation at equal intervals is carried out to sport video, so that sport video is long
Degree is n.
Preferably, feature extraction, each picture scroll are carried out to skeleton point space-time diagram using figure convolutional neural networks in S3.4
Product operation is as follows:
In formula, A is the adjacency matrix of skeleton point space-time diagram, and I is skeleton point space-time diagram from representing matrix,For the symmetric gauge laplacian decomposition of skeleton point space-time diagram, finIndicate picture scroll product input, i.e., on
One layer of graph structure data；foutIndicate the data that one layer of graph structure data obtain after this figure convolution operation, j is indicated
This picture scroll accumulates operation input data finIn number of nodes, picture scroll product number it is consistent with j；；W is the weight matrix of picture scroll product.
Preferably, the difference with standard operation is exported in S4, specially exports the difference of skeleton point and standard skeleton point coordinate
Value.This difference is measured, different evaluation is provided according to the size of difference.For nonstandard movement, according to template
The relative position of skeleton key point, to instruct trainer correctly to act.
A kind of football movement assessment system based on deep learning, including depth camera and raspberry pie processor, wherein
Depth camera is connect by data line with raspberry pie processor, includes: in raspberry pie processor
Football standard operation formulates module, the standard form of the Category criteria movement for formulating various football movements；
Video acquisition module, for obtaining the sport video for utilizing camera acquisition football training, match；
Action recognition categorization module obtains football action classification for handling the video data in sport video；
Evaluation module is matched, the mark of the football action classification in the football action classification matching criteria template for will obtain
Quasi- movement, and export the difference with standard operation.
Preferably, the raspberry pie processor is also connect with display, audio output device.
Preferably, the raspberry pie processor connects wireless communication module, connect with Ethernet.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
1. the present invention proposes a kind of human body attitude estimation model, increase LSC convolution after characteristic pattern, artificial position of introducing is believed
Breath reduces the estimation model redundancy of the human body attitude based on deep learning, in the limited situation of embedded device computing resource,
The arithmetic speed that human body attitude estimation model extracts the skeleton key point of video frame is accelerated, operation time is reduced；
2. the skeleton data with space time information is constructed using skeleton point graph structure, for limb action local in football
Movement have well characterization, discrimination improve；
3. the system integration is improved portability in an embedded device raspberry pie, solves football movement assessment
Outdoors the problem of, also reduces equipment cost and operation difficulty.
Detailed description of the invention
Fig. 1 is existing human motion recognition method schematic illustration.
Fig. 2 is a kind of football movement appraisal procedure flow chart based on deep learning.
Fig. 3 is that video length normalizes flow chart.
Fig. 4 is the three dimensional representation schematic diagram of human joint points.
Fig. 5 is that existing human body attitude estimates model structure schematic diagram.
Fig. 6 is that human body attitude of the invention estimates model structure schematic diagram.
Fig. 7 is skeleton point space-time diagram schematic diagram.
Fig. 8 is a kind of football movement assessment system schematic diagram based on deep learning.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent；
In order to better illustrate this embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent actual product
Size；
To those skilled in the art, it is to be understood that certain known features and its explanation, which may be omitted, in attached drawing
's.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
Embodiment 1
The present embodiment provides a kind of, and the football based on deep learning acts appraisal procedure, such as Fig. 2, comprising the following steps:
S1: the standard form of the Category criteria movement of various football movements is formulated；
The standard form of various football action classification standard operations is formulated in step S1, comprising the following steps:
S1.1: it collects the open video of the standard operation of various football action classifications and carries out manual sort；
S1.2: skeleton key point is sought to the standard operation of each classification using human body attitude estimation model, and right
The skeleton key point of the standard operation of each classification is averaging, and the standard operation of each classification obtains a skeleton point sequence
Column, all skeleton point Sequence composition standard forms.
Human body attitude estimates model, specific as follows:
S2: camera acquisition football training, the sport video of match are utilized；
S3: the video data in processing sport video obtains football action classification；
The video data in sport video is handled in step S3, obtains football action classification, comprising the following steps:
S3.1: sport video length normalization method is enabled；
Sport video length normalization method, such as Fig. 3 are enabled in step S3.1, comprising the following steps:
S3.1.1: enabling video desired length is n, judges sport video length for N, if N > n, sport video is carried out etc.
Interval sampling enables sport video length be equal to n；If N < n, interpolation at equal intervals is carried out to sport video, so that sport video is long
Degree is n.
S3.2: a bone point sequence, such as Fig. 4 are all obtained in each frame of sport video using human body attitude estimation model；
S3.3: being connected with each other in the skeleton point of each frame of sport video according to human physiological structure, same between different frame
The skeleton point of position is connected with each other, and is constructed skeleton point space-time diagram G=(V, E), wherein V is number of nodes, that is, bone points, indicates bone
The space structure of bone point, E are side, that is, feature vector in graph structure, indicate the connection between different frame, represent each specific section
The track of point over time, such as Fig. 7；
S3.4: feature extraction is carried out to skeleton point space-time diagram using figure convolutional neural networks, obtains the football action classification.
Feature extraction, each picture scroll product fortune are carried out to skeleton point space-time diagram using figure convolutional neural networks in step S3.4
It calculates as follows:
In formula, A is the adjacency matrix of skeleton point space-time diagram, and I is skeleton point space-time diagram from representing matrix,For the symmetric gauge laplacian decomposition of skeleton point space-time diagram, finIndicate picture scroll product input, i.e., on
One layer of graph structure data；foutIndicate the data that one layer of graph structure data obtain after this figure convolution operation, j is indicated
This picture scroll accumulates operation input data finIn number of nodes, picture scroll product number it is consistent with j；；W is the weight matrix of picture scroll product.
S4: the standard operation of the football action classification in the football action classification matching criteria template obtained with S3, and it is defeated
Out with the difference of standard operation.
The difference with standard operation is exported in step S4, specially exports the difference of skeleton point and standard skeleton point coordinate.
Human body attitude estimates model, specific as follows:
Existing method Mask R-CNN structural schematic diagram is as shown in figure 5, video frame of the input after normalizing, core network
For VGG16 convolutional neural networks；Characteristic pattern is the high dimensional feature matrix extracted after convolutional neural networks convolution, has video
The high dimensional feature of frame；RPN (Region Proposal Network) network is the Proposals extracted in video frame
Regions network；RoIAlign is to carry out further refine to the region Proposal regions, is obtained due to RPN
Proposal regions is a rough region, thus will also by RoIAlign to the coordinate of regional location carry out into
One step returns to obtain ROI (Area-of-interest, area-of-interest)；R-CNN carries out classification to each ROI and returns with position
Return, and obtains the segmentation of pixel scale to ROI deconvolution using FCN (fully connection network).Existing method
In extraction feature is individually connected entirely twice to each ROI, the port number of each ROI is consistent with characteristic pattern, Quan Lian
It connects and inner product operation is carried out to all channels, computationally intensive and weight cannot be shared, and a large amount of computing resources and time can be occupied.
In face of this problem, the present embodiment proposes a kind of human body attitude estimation model, structure as shown in Figure 6, in order to reduce
LSC (Large separable is not arranged in shared full connection, the present invention to weight to each ROI after characteristic pattern twice
Convolution) convolution, when the port number for the characteristic pattern that core network obtains is 2048, port number becomes 490 after LSC,
The port number of the ROI obtained can be fewer than the port number in Fig. 5, and the weight of LSC is subsequent R-CNN network share.It uses
PSROI pooling replaces RoIAlign, location information is manually introduced when Proposal regions carries out pond, to have
Effect improves deeper neural network to the sensitivity of object space.By the improvement of this two step, in last R-CNN sub-network
In only one layer of full connection, and the port number of ROI is reduced, and is reduced full connection inner product operation, is accelerated model calculation speed.
Embodiment 2
The present embodiment provides a kind of based on a kind of football movement appraisal procedure based on deep learning described in embodiment 1
A kind of football movement assessment system based on deep learning, such as Fig. 8, including depth camera and raspberry pie processor, wherein deep
Degree camera is connect by data line with raspberry pie processor, includes: in raspberry pie processor
Football standard operation formulates module, the standard form of the Category criteria movement for formulating various football movements；
Video acquisition module, for obtaining the sport video for utilizing camera acquisition football training, match；
Action recognition categorization module obtains football action classification for handling the video data in sport video；
Evaluation module is matched, the mark of the football action classification in the football action classification matching criteria template for will obtain
Quasi- movement, and export the difference with standard operation.
The raspberry pie processor is also connect with display, audio output device.
The raspberry pie processor connects wireless communication module, connect with Ethernet.
The same or similar label correspond to the same or similar components；
The terms describing the positional relationship in the drawings are only for illustration, should not be understood as the limitation to this patent；
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair
The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description
To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this
Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention
Protection scope within.
Claims (10)
1. a kind of football based on deep learning acts appraisal procedure, which comprises the following steps:
S1: the standard form of the Category criteria movement of various football movements is formulated；
S2: camera acquisition football training, the sport video of match are utilized；
S3: the video data in processing sport video obtains football action classification；
S4: the standard operation of the football action classification in the football action classification matching criteria template obtained with S3, and export with
The difference of standard operation.
2. the football according to claim 1 based on deep learning acts appraisal procedure, which is characterized in that made in step S1
The standard form of fixed various football action classification standard operations, comprising the following steps:
S1.1: it collects the open video of the standard operation of various football action classifications and carries out manual sort；
S1.2: skeleton key point is sought to the standard operation of each classification using human body attitude estimation model, and to each
The skeleton key point of the standard operation of classification is averaging, and the standard operation of each classification obtains a bone point sequence,
All skeleton point Sequence composition standard forms.
3. the football according to claim 2 based on deep learning acts appraisal procedure, which is characterized in that the human body appearance
State estimates model, specific as follows:
After the video frame of sport video after normalization is input to VGG16 or ResNet convolutional neural networks, obtain that there is view
The characteristic pattern of the high dimensional feature of frequency frame, while being also high dimensional feature matrix；Characteristic pattern is obtained through RPN network after LSC convolution again again
To the region Proposal regions, it is input to again after being detected using PSROI pooling to Proposal regions
R-CNN network carries out classification and position returns, and R-CNN network only has one layer of full articulamentum, finally obtains bone point sequence.
4. the football according to claim 3 based on deep learning acts appraisal procedure, which is characterized in that in step S3
The video data in sport video is managed, football action classification is obtained, comprising the following steps:
S3.1: sport video length normalization method is enabled；
S3.2: a bone point sequence is all obtained in each frame of sport video using human body attitude estimation model；
S3.3: it is connected with each other in the skeleton point of each frame of sport video according to human physiological structure, same position between different frame
Skeleton point be connected with each other, construct skeleton point space-time diagram G=(V, E), wherein V is number of nodes, that is, bone points, indicates skeleton point
Space structure, E be graph structure in side, that is, feature vector, indicate different frame between connection, represent each specific node with
The track of time passage；
S3.4: feature extraction is carried out to skeleton point space-time diagram using figure convolutional neural networks, obtains the football action classification.
5. the football according to claim 4 based on deep learning acts appraisal procedure, which is characterized in that in step S3.1
Enable sport video length normalization method, comprising the following steps:
S3.1.1: enabling video desired length is n, judges sport video length for N, if N > n, carries out at equal intervals to sport video
Sampling enables sport video length be equal to n；If N < n, interpolation at equal intervals is carried out to sport video, so that sport video length is
n。
6. the football according to claim 5 based on deep learning acts appraisal procedure, which is characterized in that in step S3.4
Feature extraction is carried out to skeleton point space-time diagram using figure convolutional neural networks, each figure convolution algorithm is as follows:
In formula, A is the adjacency matrix of skeleton point space-time diagram, and I is skeleton point space-time diagram from representing matrix,
For the symmetric gauge laplacian decomposition of skeleton point space-time diagram, finIndicate the input of picture scroll product, i.e., upper one layer of graph structure number
According to；foutIndicate the data that one layer of graph structure data obtain after this figure convolution operation, j indicates this figure convolution operation
Input data finIn number of nodes, picture scroll product number it is consistent with j；；W is the weight matrix of picture scroll product.
7. the football according to claim 6 based on deep learning acts appraisal procedure, which is characterized in that defeated in step S4
Out with the difference of standard operation, the difference of skeleton point and standard skeleton point coordinate is specially exported.
8. a kind of foot based on deep learning of the football movement appraisal procedure according to claim 7 based on deep learning
Ball acts assessment system, which is characterized in that including depth camera and raspberry pie processor, wherein depth camera passes through number
It is connect according to line with raspberry pie processor, includes: in raspberry pie processor
Football standard operation formulates module, the standard form of the Category criteria movement for formulating various football movements；Video is adopted
Collect module, for obtaining the sport video for utilizing camera acquisition football training, match；
Action recognition categorization module obtains football action classification for handling the video data in sport video；
Evaluation module is matched, the standard of the football action classification is moved in the football action classification matching criteria template for will obtain
Make, and exports the difference with standard operation.
9. the football according to claim 8 based on deep learning acts assessment system, which is characterized in that the raspberry
Processor is sent also to connect with display, audio output device.
10. the football based on deep learning according to claim 8 or claim 9 acts assessment system, which is characterized in that described
Raspberry pie processor connects wireless communication module, connect with Ethernet.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910143782.7A CN109948459B (en) | 2019-02-25 | 2019-02-25 | Football action evaluation method and system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910143782.7A CN109948459B (en) | 2019-02-25 | 2019-02-25 | Football action evaluation method and system based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109948459A true CN109948459A (en) | 2019-06-28 |
| CN109948459B CN109948459B (en) | 2023-08-25 |
Family
ID=67006989
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910143782.7A Active CN109948459B (en) | 2019-02-25 | 2019-02-25 | Football action evaluation method and system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109948459B (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110633736A (en) * | 2019-08-27 | 2019-12-31 | 电子科技大学 | Human body falling detection method based on multi-source heterogeneous data fusion |
| CN110705418A (en) * | 2019-09-25 | 2020-01-17 | 西南大学 | Taekwondo kicking motion video capture and scoring system based on deep LabCut |
| CN110929242A (en) * | 2019-11-20 | 2020-03-27 | 上海交通大学 | Method and system for carrying out attitude-independent continuous user authentication based on wireless signals |
| CN110941990A (en) * | 2019-10-22 | 2020-03-31 | 泰康保险集团股份有限公司 | Method and device for evaluating human body actions based on skeleton key points |
| CN111223549A (en) * | 2019-12-30 | 2020-06-02 | 华东师范大学 | Mobile end system and method for disease prevention based on posture correction |
| CN111259796A (en) * | 2020-01-16 | 2020-06-09 | 东华大学 | Lane line detection method based on image geometric features |
| CN111460960A (en) * | 2020-03-27 | 2020-07-28 | 重庆电政信息科技有限公司 | Motion classification and counting method |
| CN111507192A (en) * | 2020-03-19 | 2020-08-07 | 北京捷通华声科技股份有限公司 | Appearance instrument monitoring method and device |
| CN111666844A (en) * | 2020-05-26 | 2020-09-15 | 电子科技大学 | Badminton player motion posture assessment method |
| CN111680560A (en) * | 2020-05-07 | 2020-09-18 | 南通大学 | Pedestrian re-identification method based on space-time characteristics |
| CN111881859A (en) * | 2020-07-31 | 2020-11-03 | 北京融链科技有限公司 | Template generation method and device |
| CN111986260A (en) * | 2020-09-04 | 2020-11-24 | 北京小狗智能机器人技术有限公司 | Image processing method and device and terminal equipment |
| CN112069979A (en) * | 2020-09-03 | 2020-12-11 | 浙江大学 | Real-time action recognition man-machine interaction system |
| CN112370045A (en) * | 2020-10-15 | 2021-02-19 | 北京大学 | Functional action detection method and system based on artificial intelligence |
| CN112749684A (en) * | 2021-01-27 | 2021-05-04 | 萱闱(北京)生物科技有限公司 | Cardiopulmonary resuscitation training and evaluating method, device, equipment and storage medium |
| CN112818800A (en) * | 2021-01-26 | 2021-05-18 | 中国人民解放军火箭军工程大学 | Physical exercise evaluation method and system based on human skeleton point depth image |
| CN113052061A (en) * | 2021-03-22 | 2021-06-29 | 中国石油大学(华东) | Speed skating athlete motion identification method based on human body posture estimation |
| CN113095248A (en) * | 2021-04-19 | 2021-07-09 | 中国石油大学(华东) | Technical action correction method for badminton |
| CN113537128A (en) * | 2021-07-29 | 2021-10-22 | 广州中金育能教育科技有限公司 | Method, system and equipment for comparing and analyzing continuous actions based on deep learning posture assessment |
| CN113642525A (en) * | 2021-09-02 | 2021-11-12 | 浙江大学 | Infant neural development assessment method and system based on skeletal points |
| CN114091511A (en) * | 2021-09-22 | 2022-02-25 | 广东工业大学 | Fitness action scoring method, system and device based on space-time information |
| CN114821812A (en) * | 2022-06-24 | 2022-07-29 | 西南石油大学 | Deep learning-based skeleton point action recognition method for pattern skating players |
| CN115661929A (en) * | 2022-10-28 | 2023-01-31 | 北京此刻启动科技有限公司 | Time sequence feature coding method and device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103390174A (en) * | 2012-05-07 | 2013-11-13 | 深圳泰山在线科技有限公司 | Physical education assisting system and method based on human body posture recognition |
| CN107220608A (en) * | 2017-05-22 | 2017-09-29 | 华南理工大学 | What a kind of basketball action model was rebuild and defended instructs system and method |
| CN108664896A (en) * | 2018-04-16 | 2018-10-16 | 彭友 | Fencing action acquisition methods based on OpenPose and computer storage media |
| CN108960078A (en) * | 2018-06-12 | 2018-12-07 | 温州大学 | A method of based on monocular vision, from action recognition identity |
-
2019
- 2019-02-25 CN CN201910143782.7A patent/CN109948459B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103390174A (en) * | 2012-05-07 | 2013-11-13 | 深圳泰山在线科技有限公司 | Physical education assisting system and method based on human body posture recognition |
| CN107220608A (en) * | 2017-05-22 | 2017-09-29 | 华南理工大学 | What a kind of basketball action model was rebuild and defended instructs system and method |
| CN108664896A (en) * | 2018-04-16 | 2018-10-16 | 彭友 | Fencing action acquisition methods based on OpenPose and computer storage media |
| CN108960078A (en) * | 2018-06-12 | 2018-12-07 | 温州大学 | A method of based on monocular vision, from action recognition identity |
Non-Patent Citations (1)
| Title |
|---|
| 万晓依: "《基于时空结构关系的3D人体行为识别研究》", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110633736A (en) * | 2019-08-27 | 2019-12-31 | 电子科技大学 | Human body falling detection method based on multi-source heterogeneous data fusion |
| CN110705418B (en) * | 2019-09-25 | 2021-11-30 | 西南大学 | Taekwondo kicking motion video capture and scoring system based on deep LabCut |
| CN110705418A (en) * | 2019-09-25 | 2020-01-17 | 西南大学 | Taekwondo kicking motion video capture and scoring system based on deep LabCut |
| CN110941990A (en) * | 2019-10-22 | 2020-03-31 | 泰康保险集团股份有限公司 | Method and device for evaluating human body actions based on skeleton key points |
| CN110941990B (en) * | 2019-10-22 | 2023-06-16 | 泰康保险集团股份有限公司 | Method and device for evaluating human body actions based on skeleton key points |
| CN110929242A (en) * | 2019-11-20 | 2020-03-27 | 上海交通大学 | Method and system for carrying out attitude-independent continuous user authentication based on wireless signals |
| CN110929242B (en) * | 2019-11-20 | 2020-07-10 | 上海交通大学 | Method and system for carrying out attitude-independent continuous user authentication based on wireless signals |
| CN111223549A (en) * | 2019-12-30 | 2020-06-02 | 华东师范大学 | Mobile end system and method for disease prevention based on posture correction |
| CN111259796A (en) * | 2020-01-16 | 2020-06-09 | 东华大学 | Lane line detection method based on image geometric features |
| CN111507192A (en) * | 2020-03-19 | 2020-08-07 | 北京捷通华声科技股份有限公司 | Appearance instrument monitoring method and device |
| CN111460960A (en) * | 2020-03-27 | 2020-07-28 | 重庆电政信息科技有限公司 | Motion classification and counting method |
| CN111680560A (en) * | 2020-05-07 | 2020-09-18 | 南通大学 | Pedestrian re-identification method based on space-time characteristics |
| CN111666844A (en) * | 2020-05-26 | 2020-09-15 | 电子科技大学 | Badminton player motion posture assessment method |
| CN111881859A (en) * | 2020-07-31 | 2020-11-03 | 北京融链科技有限公司 | Template generation method and device |
| CN112069979A (en) * | 2020-09-03 | 2020-12-11 | 浙江大学 | Real-time action recognition man-machine interaction system |
| CN112069979B (en) * | 2020-09-03 | 2024-02-02 | 浙江大学 | Real-time action recognition man-machine interaction system |
| CN111986260A (en) * | 2020-09-04 | 2020-11-24 | 北京小狗智能机器人技术有限公司 | Image processing method and device and terminal equipment |
| CN112370045A (en) * | 2020-10-15 | 2021-02-19 | 北京大学 | Functional action detection method and system based on artificial intelligence |
| CN112818800A (en) * | 2021-01-26 | 2021-05-18 | 中国人民解放军火箭军工程大学 | Physical exercise evaluation method and system based on human skeleton point depth image |
| CN112749684A (en) * | 2021-01-27 | 2021-05-04 | 萱闱(北京)生物科技有限公司 | Cardiopulmonary resuscitation training and evaluating method, device, equipment and storage medium |
| CN113052061A (en) * | 2021-03-22 | 2021-06-29 | 中国石油大学(华东) | Speed skating athlete motion identification method based on human body posture estimation |
| CN113095248A (en) * | 2021-04-19 | 2021-07-09 | 中国石油大学(华东) | Technical action correction method for badminton |
| CN113537128A (en) * | 2021-07-29 | 2021-10-22 | 广州中金育能教育科技有限公司 | Method, system and equipment for comparing and analyzing continuous actions based on deep learning posture assessment |
| CN113642525A (en) * | 2021-09-02 | 2021-11-12 | 浙江大学 | Infant neural development assessment method and system based on skeletal points |
| CN114091511A (en) * | 2021-09-22 | 2022-02-25 | 广东工业大学 | Fitness action scoring method, system and device based on space-time information |
| CN114091511B (en) * | 2021-09-22 | 2024-08-02 | 广东工业大学 | Body-building action scoring method, system and device based on space-time information |
| CN114821812A (en) * | 2022-06-24 | 2022-07-29 | 西南石油大学 | Deep learning-based skeleton point action recognition method for pattern skating players |
| CN114821812B (en) * | 2022-06-24 | 2022-09-13 | 西南石油大学 | Deep learning-based skeleton point action recognition method for pattern skating players |
| CN115661929A (en) * | 2022-10-28 | 2023-01-31 | 北京此刻启动科技有限公司 | Time sequence feature coding method and device, electronic equipment and storage medium |
| CN115661929B (en) * | 2022-10-28 | 2023-11-17 | 北京此刻启动科技有限公司 | Time sequence feature coding method and device, electronic equipment and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109948459B (en) | 2023-08-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109948459A (en) | A kind of football movement appraisal procedure and system based on deep learning | |
| CN112861624A (en) | Human body posture detection method, system, storage medium, equipment and terminal | |
| CN107944431B (en) | A kind of intelligent identification Method based on motion change | |
| CN105243386B (en) | Face living body judgment method and system | |
| WO2019054638A1 (en) | Image analysis method and apparatus and computer program | |
| CN110738154A (en) | pedestrian falling detection method based on human body posture estimation | |
| CN111437583A (en) | Badminton basic action auxiliary training system based on Kinect | |
| CN110490109B (en) | Monocular vision-based online human body rehabilitation action recognition method | |
| CN113239797B (en) | Human body action recognition method, device and system | |
| CN113111767A (en) | Fall detection method based on deep learning 3D posture assessment | |
| US20200279102A1 (en) | Movement monitoring system | |
| CN110929687A (en) | Multi-user behavior recognition system based on key point detection and working method | |
| CN109492534A (en) | A kind of pedestrian detection method across scene multi-pose based on Faster RCNN | |
| CN111833439B (en) | Artificial intelligence based ammunition throwing analysis and mobile simulation training method | |
| CN116092199A (en) | Employee working state identification method and identification system | |
| Li et al. | Automated fine motor evaluation for developmental coordination disorder | |
| Zhi-chao et al. | Key pose recognition toward sports scene using deeply-learned model | |
| CN106846372A (en) | Human motion quality visual A+E system and method | |
| CN108875448B (en) | Pedestrian re-identification method and device | |
| CN110532838A (en) | Object test equipment and method and storage medium | |
| Almasi et al. | Human action recognition through the first-person point of view, case study two basic task | |
| CN114187664B (en) | Rope skipping counting system based on artificial intelligence | |
| CN102148919B (en) | Method and system for detecting balls | |
| CN109993116A (en) | A kind of pedestrian mutually learnt based on skeleton recognition methods again | |
| CN113255462A (en) | Gait scoring method, system, computer program product and readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |