CN110490064A - Processing method, device, computer equipment and the computer storage medium of sports video data - Google Patents
Processing method, device, computer equipment and the computer storage medium of sports video data Download PDFInfo
- Publication number
- CN110490064A CN110490064A CN201910624923.7A CN201910624923A CN110490064A CN 110490064 A CN110490064 A CN 110490064A CN 201910624923 A CN201910624923 A CN 201910624923A CN 110490064 A CN110490064 A CN 110490064A
- Authority
- CN
- China
- Prior art keywords
- sports video
- video data
- technical movements
- sports
- data
- 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
- 238000003672 processing method Methods 0.000 title claims abstract description 16
- 238000012545 processing Methods 0.000 claims abstract description 102
- 230000009471 action Effects 0.000 claims abstract description 45
- 238000000034 method Methods 0.000 claims abstract description 27
- 238000013136 deep learning model Methods 0.000 claims abstract description 15
- 238000005516 engineering process Methods 0.000 claims description 59
- 230000020509 sex determination Effects 0.000 claims description 13
- 238000004590 computer program Methods 0.000 claims description 11
- 238000012216 screening Methods 0.000 claims description 8
- 238000010276 construction Methods 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 5
- 238000013135 deep learning Methods 0.000 claims description 2
- 238000012549 training Methods 0.000 description 11
- 238000010586 diagram Methods 0.000 description 8
- 230000003068 static effect Effects 0.000 description 5
- 230000002860 competitive effect Effects 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000004458 analytical method Methods 0.000 description 3
- 238000004891 communication Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 241001269238 Data Species 0.000 description 2
- 230000004913 activation Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/49—Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/20—Movements or behaviour, e.g. gesture recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Psychiatry (AREA)
- Social Psychology (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
This application discloses a kind of processing method of sports video data, device and computer storage mediums, are related to technical field of data processing, can more fully be analyzed sports video data, improve the precision of the processing of sports video data.It include the sports video picture for carrying technical movements label in the sports video sample data the described method includes: obtaining sports video sample data;The sports video picture for carrying technical movements label in the sports video sample data is input in deep learning model and is trained, action recognition model is constructed;When the processing for receiving sports video data is requested, the sports video picture in sports video data to be processed is input to the action recognition model, obtains the technical movements in sports video picture;Based on the solicited message carried in the processing request, the technical movements in each sports video picture are counted, the processing result of sports video data is obtained.
Description
Technical field
The present invention relates to technical field of data processing, particularly with regard to the processing method of sports video data, device, meter
Calculate machine equipment and computer storage medium.
Background technique
Science and technology is changing people's lives, especially present big data industry, and big data is brought to certain industries
Earth-shaking variation.For example, the sport industry such as sport industry, either football, basketball or racing car, utilizes big data
The functions such as capture, storage, analysis make these fields that great change have occurred.
General competitive sports company can all be absorbed in the analysis of race data, the technical team of core can by every game into
Row second grade data slicer, and various dimensions processing is carried out to the sports video data in each slice, for example, for Basketball Match,
The number that different technologies movement occurs in sports video data is counted, for football match, is acted based on sportsman's difference to body
Video data is educated to classify, in order to for user push it is more personalized, more meet the sports video data of hobby, improve use
The experience of family viewing sports video.
Generally use in the prior art be artificial statistics or artificial editing mode come to sports video data into
Row processing, needs to expend a large amount of manpowers, in addition, the different obtained data results of artificial treatment make there are certain subjectivity
The consistency for obtaining data processed result is poor.
Summary of the invention
In view of this, the present invention provides a kind of processing method of sports video data, device, computer equipment and calculating
Machine storage medium, main purpose are to solve at present since the obtained data result of artificial treatment has certain subjectivity,
The problem for causing the processing result consistency of sports video data poor.
According to the present invention on one side, a kind of processing method of sports video data is provided, this method comprises:
Sports video sample data is obtained, includes the body for carrying technical movements label in the sports video sample data
Educate video pictures;
The sports video picture that technical movements label is carried in the sports video sample data is input to depth
It practises and being trained in model, construct action recognition model, the action recognition model skill in sports video picture for identification
Art movement;
When the processing for receiving sports video data is requested, by the sports video picture in sports video data to be processed
It is input to the action recognition model, obtains the technical movements in sports video picture;
Based on the solicited message carried in the processing request, unite to the technical movements in each sports video picture
Meter, obtains the processing result of sports video data.
Further, the deep learning model is depth residual error network, and the depth residual error network includes multilayered structure,
It is described that the sports video picture that technical movements label is carried in the sports video sample data is input to deep learning mould
It is trained in type, building action recognition model includes:
The feature that sports video picture is extracted by the convolutional layer of the depth residual error network, obtains sports video picture and exists
Characteristic parameter in different technologies movement;
Feature by the pond layer of the depth residual error network to the sports video picture in different technologies movement
Parameter carries out dimension-reduction treatment, obtains characteristic parameter of the sports video picture in different technologies movement after dimension-reduction treatment;
Sports video picture is in different skills after summarizing the dimension-reduction treatment by the full articulamentum of the depth residual error network
Characteristic parameter in art movement obtains the weighted value that different technologies movement is characterized in sports video picture;
By the classification layer of the depth residual error network according to characterization different technologies movement in the sports video picture
Weighted value generates the recognition result of technical movements in sports video picture.
Further, the acquisition sports video sample data, specifically includes:
The sports video data for collecting acquisition equipment capture carries out the sports video data according to prefixed time interval
Segmentation, obtains multistage sports video frame data;
It is educated from every segment body and chooses multiple sports video pictures in video requency frame data, it is dynamic based on the technology in sports video picture
The sports video picture of opposing is marked, and obtains sports video sample data.
Further, when the processing request is counts the frequency of occurrence of the first technical movements in sports video data,
The first technical movements of request statistics are carried in the solicited message, it is described based on the request letter carried in the processing request
Breath, counts the technical movements in each sports video picture, the processing result for obtaining sports video data includes:
By carrying out continuous sex determination to the technical movements identified in sports video data, sports video data is counted
In each technical movements frequency of occurrence;
According to the frequency of occurrence of technical movements each in the sports video data, determine the first technical movements goes out occurrence
Number, obtains the processing result of sports video data;
It is described by carrying out continuous sex determination to the technical movements that identify in sports video data, count sports video
The frequency of occurrence of each technical movements in data, specifically includes:
Based on the technical movements that video pictures each in sports video data identify, if connected in sports video data
The technical movements that the sports video picture recognition that continuous appearance is greater than or equal to preset quantity obtains are identical, then determine sports video number
According to technical movements of middle appearance, the frequency of occurrence of each technical movements in sports video data is counted.
Further, described when the processing for receiving sports video data is requested, by sports video data to be processed
In sports video picture be input to the action recognition model, it is described after obtaining the technical movements in sports video picture
Method further include:
The sports video data of different angle and different distance based on acquisition equipment capture, to the sports video number
According to being divided, the sports video data of different acquisition angle and different acquisition distance is obtained.
Further, the sports video data of the different angle and different distance based on acquisition equipment capture is right
The sports video data is divided, and is specifically included:
The human body target in sport video pictures is identified by human testing algorithm, determines human body target in sports video figure
Coordinate and size in piece;
According to coordinate and size of the human body target in sports video picture, human body target is calculated in sports video
Accounting value in picture;
Accounting value based on human body target in sports video picture divides the sports video data, obtains not
With the sports video data of acquisition angles and different acquisition distance.
Further, when the processing request is screens the wonderful in sports video data, the solicited message
In carry corresponding second technical movements of wonderful, it is described based on the solicited message carried in the processing request, to every
Technical movements in a sport video pictures are counted, and the processing result for obtaining sports video data includes:
Based on the corresponding acquisition angles of sports video data and acquisition distance, acquisition angles are filtered out as forward direction and are adopted
The sports video data of collection distance within a preset range is as candidate wonderful;
By carrying out continuous sex determination to the technical movements identified in candidate wonderful, candidate wonderful is counted
The technical movements of middle appearance;
According to the technical movements occurred in the candidate wonderful, determine the excellent candidate of the second technical movements occur
Section, obtains the processing result of sports video data;
It is described by carrying out continuous sex determination to the technical movements that identify in candidate wonderful, statistics is candidate excellent
The technical movements occurred in segment, specifically include:
Based on the technical movements that each video pictures in candidate wonderful identify, if connected in candidate wonderful
The technical movements that the sports video picture recognition that continuous appearance is greater than or equal to preset quantity obtains are identical, then determine excellent candidate
Occur a technical movements in section, counts the technical movements occurred in candidate wonderful.
According to the present invention on the other hand, a kind of processing unit of sports video data is provided, described device includes:
Acquiring unit includes to carry skill in the sports video sample data for obtaining sports video sample data
The sports video picture of art movement label;
Construction unit, for the sports video picture of technical movements label will to be carried in the sports video sample data
It is input in deep learning model and is trained, construct action recognition model, sport regards the action recognition model for identification
Technical movements in frequency picture;
Recognition unit, for when receive sports video data processing request when, will be in sports video data to be processed
Sports video picture be input to the action recognition model, obtain the technical movements in sports video picture;
Statistic unit, the solicited message for being carried in being requested based on the processing, in each sports video picture
Technical movements are counted, and the processing result of sports video data is obtained.
Further, the deep learning model is depth residual error network, and the depth residual error network includes multilayered structure,
The construction unit includes:
Extraction module extracts the feature of sports video picture for the convolutional layer by the depth residual error network, obtains
Characteristic parameter of the sports video picture in different technologies movement;
Dimensionality reduction module, for the pond layer by the depth residual error network to the sports video picture in different technologies
Characteristic parameter in movement carries out dimension-reduction treatment, obtains feature of the sports video picture in different technologies movement after dimension-reduction treatment
Parameter;
Summarizing module, for summarizing sports video after the dimension-reduction treatment by the full articulamentum of the depth residual error network
Characteristic parameter of the picture in different technologies movement obtains the weighted value that different technologies movement is characterized in sports video picture;
Generation module characterizes not for the classification layer by the depth residual error network according in the sports video picture
The recognition result of technical movements in sports video picture is generated with the weighted value of technical movements.
Further, the acquiring unit includes:
Collection module, for collecting the sports video data of acquisition equipment capture, according to prefixed time interval to the body
It educates video data to be split, obtains multistage sports video frame data;
Mark module chooses multiple sports video pictures for educating from every segment body in video requency frame data, be based on sports video
The sports video picture is marked in technical movements in picture, obtains sports video sample data.
Further, when the processing request is counts the frequency of occurrence of the first technical movements in sports video data,
The first technical movements of request statistics are carried in the solicited message, the statistic unit includes:
First statistical module, for being sentenced by carrying out continuity to the technical movements identified in sports video data
It is fixed, count the frequency of occurrence of each technical movements in sports video data;
First determining module determines for the frequency of occurrence according to technical movements each in the sports video data
The frequency of occurrence of one technical movements obtains the processing result of sports video data;
First statistical module, specifically for the technology identified based on video pictures each in sports video data
Movement, if continuously occurring being greater than or equal to the technology that the sports video picture recognition of preset quantity obtains in sports video data
It acts identical, then determines occur a technical movements in sports video data, count each technical movements in sports video data
Frequency of occurrence.
Further, described device further include:
Division unit, for it is described when receive sports video data processing request when, by sports video to be processed
Sports video picture in data is input to the action recognition model, after obtaining the technical movements in sports video picture,
The sports video data of different angle and different distance based on acquisition equipment capture, draws the sports video data
Point, obtain the sports video data of different acquisition angle and different acquisition distance.
Further, the division unit includes:
Identification module determines human body mesh for identifying the human body target in sport video pictures by human testing algorithm
The coordinate and size being marked in sports video picture;
Computing module calculates human body for the coordinate and size according to the human body target in sports video picture
Accounting value of the target in sports video picture;
Division module, for based on accounting value of the human body target in sports video picture to the sports video data into
Row divides, and obtains the sports video data of different acquisition angle and different acquisition distance.
Further, when the processing request is screens the wonderful in sports video data, the solicited message
In carry corresponding second technical movements of wonderful, the statistic unit includes:
Screening module, for filtering out acquisition angle based on the corresponding acquisition angles of sports video data and acquisition distance
Degree is for positive and acquisition apart from sports video data within a preset range as candidate wonderful;
Second statistical module, for being sentenced by carrying out continuity to the technical movements identified in candidate wonderful
It is fixed, count the technical movements occurred in candidate wonderful;
Second determining module, for determining the second skill occur according to the technical movements occurred in the candidate wonderful
The candidate wonderful of art movement, obtains the processing result of sports video data;
Second statistical module, specifically for the technology identified based on each video pictures in candidate wonderful
Movement, if continuously occurring being greater than or equal to the technology that the sports video picture recognition of preset quantity obtains in candidate wonderful
It acts identical, then determines occur a technical movements in candidate wonderful, it is dynamic to count the technology occurred in candidate wonderful
Make.
Another aspect according to the present invention provides a kind of computer equipment, including memory and processor, the storage
Device is stored with computer program, and the processor realizes the processing method of sports video data when executing the computer program
Step.
Another aspect according to the present invention provides a kind of computer storage medium, is stored thereon with computer program, institute
The step of stating the processing method that sports video data is realized when computer program is executed by processor.
By above-mentioned technical proposal, the present invention provides a kind of processing method and processing device of sports video data, passes through acquisition
Sports video sample data, the sports video picture for carrying technical movements label for including by sports video sample data input
It is trained into deep learning model, action recognition model is constructed, thus according to action recognition model come to sports video number
According to being handled, without manually being counted to sports video data, it can automatically identify and occur in sports video data
Technical movements improve the processing accuracy of sports video data.With the side for using artificial statistics or artificial editing in the prior art
Formula compares the mode handled sports video data, and the present invention identifies sport by the technology identification model of building
Technical movements in video data are not necessarily to manual identified or artificial editing, the processing time of sports data are saved, due to artificial
There are certain subjectivity, the processing results of the sports video data based on the generation of technology identification model to have higher essence for processing
Degree improves the flexibility of sports video data processing.
Detailed description of the invention
By reading the following detailed description of the preferred embodiment, various other advantages and benefits are common for this field
Technical staff will become clear.The drawings are only for the purpose of illustrating a preferred embodiment, and is not considered as to the present invention
Limitation.And throughout the drawings, the same reference numbers will be used to refer to the same parts.In the accompanying drawings:
Fig. 1 shows a kind of flow diagram of the processing method of sports video data provided in an embodiment of the present invention;
Fig. 2 shows the flow diagrams of the processing method of another sports video data provided in an embodiment of the present invention;
Fig. 3 shows a kind of structural schematic diagram of the processing unit of sports video data provided in an embodiment of the present invention;
Fig. 4 shows the structural schematic diagram of the processing unit of another sports video data provided in an embodiment of the present invention;
Fig. 5 shows the structural schematic diagram of the processing unit of another sports video data provided in an embodiment of the present invention.
Specific embodiment
Exemplary embodiments of the present disclosure are described in more detail below with reference to accompanying drawings.Although showing the disclosure in attached drawing
Exemplary embodiment, it being understood, however, that may be realized in various forms the disclosure without should be by embodiments set forth here
It is limited.On the contrary, these embodiments are provided to facilitate a more thoroughly understanding of the present invention, and can be by the scope of the present disclosure
It is fully disclosed to those skilled in the art.
The embodiment of the invention provides a kind of processing methods of sports video data, can carry out more to sports video data
Comprehensive analysis, improves the precision of the processing of sports video data, as shown in Figure 1, this method comprises:
101, sports video sample data is obtained.
It wherein, include the sports video picture for carrying technical movements label, body here in sports video sample data
All kinds of sports video datas such as educate video sample data and may include but be not limited to basketball, football, shuttlecock, and for not
Different technical movements can be accordingly set with ball, for example, acting for basketball technique, 13 classifications can be set, specifically
It including 12 technical movements and 1 static movement, such as backboard, assists, grab, dunk shot, football technique being acted, Ke Yishe
11 classifications are set, 10 technical movements and 1 static movement, such as corner-kick, free kick, foul ball are specifically included.
In the present embodiment, sports video sample data is the form of sports video picture, can specifically be collected into body
After educating video data, by being split to sports video data, multistage video requency frame data is obtained, in each video requency frame data
Comprising multiple sports video pictures, to choose multiple sports video pictures as sports video sample data.
102, the sports video picture that technical movements label is carried in the sports video sample data is input to depth
It is trained in degree learning model, constructs action recognition model.
Wherein, the action recognition model technical movements in sports video picture for identification, since deep learning model has
There are mapping relations between trained sports video picture and technical movements, for the sports video figure being input in model
Piece can identify the technical movements in sport video pictures by the action recognition model of building, certainly in sports video
The picture of non-generation technique movement can identify static movement, and the property of may also adapt to increases new technical movements, here
To the classifications of technical movements without limiting.
For the embodiment of the present invention, deep learning model here can be depth residual error network model, can specifically lead to
Repetition training sports video sample data is crossed to construct the network structure of action recognition model, which can be to input
Sports video picture is trained, and provides correct Input output Relationship, that is, is directed to sports video picture in different technologies
Can all there be the recognition result accordingly exported in movement, that is, there are the probability feelings of different technologies movement in sports video picture
Condition.
The structure of specific depth residual error network model can pass through convolutional layer, pond layer, full articulamentum and classification layer knot
Structure realizes that convolutional layer here is equivalent to the hidden layer of depth residual error network, can be multilayered structure, for extracting deeper time
Sports video picture different technologies movement on characteristic parameter;In depth residual error network model, in order to reduce parameter, subtract
Low calculating, usually pond layer is inserted at interval in continuous convolution layer;Here full articulamentum is similar to convolutional layer, the mind of convolutional layer
It is connected through member with upper one layer of output regional area, of course for reducing, output feature vector is excessive, and two full connections can be set
Layer integrates the characteristic of training output after training data is by the training of several convolutional layers.
103, when the processing for receiving sports video data is requested, by the sports video in sports video data to be processed
Picture is input to the action recognition model, obtains the technical movements in sports video picture.
Wherein, the sports video number that processing requests the sports video data to be processed of processing usually to acquire in the recent period
According to, and acted in sports video data comprising multiple technologies, it can identify that sport to be processed regards by action recognition model
Frequency technical movements included in, frequency of occurrence and the technology that can also further count every kind of technical movements certainly are dynamic
Make the information such as time of occurrence.
It should be noted that sports video data to be processed is video stream data, before being input to action recognition model, together
Sample needs to pre-process sports video data to be processed, and obtained sports video picture is input in action recognition model
It is identified.
104, based on the solicited message that carries in the processing request, to the technical movements in each sports video picture into
Row statistics, obtains the processing result of sports video data.
In embodiments of the present invention, processing request is specifically as follows a certain technical movements in request statistics sports video data
Frequency of occurrence, can also for request screening sports video data in wonderful, here to processing request particular content
Without limiting.
It is dynamic specifically the technology in each sports video picture can be summarized according to the solicited message carried in processing request
Make, and is screened out from it the recognition result for requesting corresponding technical movements, the processing result as sports video data.For example, place
Reason request is the frequency of occurrence of dunk shot technical movements, by uniting to occurring dunk shot technical movements obtained by action recognition model
Meter obtains the number that dunk shot technical movements occur in sports video data.
It should be noted that for the technical movements identified in sports video picture, if in adjacent sports video
Only occur the primary technical movements in picture, then illustrate may not occur the technical movements in sports video data really,
But the similar movement made of sportsman or feinting deception, it here can be by being identified in multiple continuous sport video pictures
After the identical technical movements arrived, just determines the technical movements occur in sports video picture, can guarantee technical movements in this way
Statistical accuracy, and then improve the processing result of sports video data.
The embodiment of the present invention provides a kind of processing method of sports video data, by obtaining sports video sample data,
The sports video picture for carrying technical movements label for including by sports video sample data is input in deep learning model
It is trained, constructs action recognition model, to handle according to action recognition model sports video data, be not necessarily to people
Work counts sports video data, can automatically identify the technical movements occurred in sports video data, improves sport
The processing accuracy of video data.With in the prior art by the way of artificial statistics or artificial editing come to sports video number
It is compared according to the mode handled, the present invention identifies that the technology in sports video data is dynamic by the technology identification model of building
Make, is not necessarily to manual identified or artificial editing, the processing time of sports data is saved, since there are certain subjectivities for artificial treatment
Property, the processing result of the sports video data based on the generation of technology identification model has higher precision, improves sports video number
According to the flexibility of processing.
The embodiment of the invention provides the processing methods of another sports video data, can carry out to sports video data
It more fully analyzes, improves the precision of the processing of sports video data, as shown in Figure 2, which comprises
201, the sports video data for collecting acquisition equipment capture, according to prefixed time interval to the sports video data
It is split, obtains multistage sports video frame data.
Wherein, acquisition equipment is the video camera that can acquire video, certainly can also be using other with video acquisition function
The equipment of energy, here without limiting.Under normal conditions, it can all be set in different direction and different distance at competitive sports scene
Multiple acquisition equipment are set, in order to which viewer can from different perspectives and different distance preferably watches competitive sports, certainly
The process of sports tournament can also be retained, to play back the wonderful of competitive sports.
It should be noted that the interval frame of segmentation sports video data can be set here, for example, a sports video data
With 300 frames, which is divided into 3 sections, it includes 100 frames that every segment body, which educates video data, and each frame is one
Static sports video picture chooses multiple sports video pictures as sports video sample data, here to the number of segment of segmentation
Without limiting, can specifically be determined according to the accuracy requirement in realistic model training process.
202, it is educated from every segment body and chooses multiple sports video pictures in video requency frame data, based on the skill in sports video picture
The sports video picture is marked in art movement, obtains sports video sample data.
In embodiments of the present invention, due to that can include a large amount of sports video picture in sports video data, and every individual
Educating the technical movements occurred in video pictures may be different, for example, the technology occurred in the 1-5 sport video pictures is dynamic
As static, the calculating action that the 6-8 sport video pictures occurs is secondary attack, is occurred in the 9-10 sport video pictures
Technical movements are dunk shot etc., are needed before training sports video picture, based on the technical movements in sports video picture to body
It educates video pictures to be marked, to guarantee that training obtains the accuracy of action recognition model.
203, the sports video picture that technical movements label is carried in the sports video sample data is input to depth
It is trained in degree learning model, constructs action recognition model.
In order to guarantee that the normalization of the video pictures of input needs before being trained to deep learning model to mark
Sports video picture after note is initialized, for example, normalizing to [0,1] for each sports video picture, subtracting mean value
(being denoted as m) zooms to the size of 224*224 divided by standard deviation (being denoted as u) to get x=(x-m)/u, m=0.5, u=0.23 is arrived.
Wherein, depth residual error network model can be rsnet50 model, using TSN frame, in order to improve depth residual error net
The identification of network model can train benchmark of the resnet50 classifier as training performance, quite on ImagNet data set
In the pre-training to rsnet50 model.
Specifically, depth residual error network can be 4 layers of structure composition, and first layer structure is convolutional layer structure, the convolutional layer
Structure can be made of multiple convolutional layers, and the feature of sports video picture is extracted by convolutional layer, sports video picture is obtained and exists
Characteristic parameter in different technologies movement;Second layer structure is pond layer, by pond layer to sports video picture in different skills
Characteristic parameter in art movement carries out dimension-reduction treatment, obtains spy of the sports video picture in different technologies movement after dimension-reduction treatment
Levy parameter;Third layer structure is full articulamentum, and sports video picture is in different technologies after summarizing dimension-reduction treatment by full articulamentum
Characteristic parameter in movement obtains the weighted value that different technologies movement is characterized in sports video picture;Four-layer structure is classification
Layer generates technology in sports video picture according to the weighted value for characterizing different technologies movement in sports video picture by classification layer
The recognition result of movement constructs action recognition model.
For example, convolutional neural networks include 13 convolutional layers, each convolutional layer is can be set in 3 full articulamentums here
Convolution kernel number is respectively 64,64,128,128,256,256,512,512,512,512,512,512, and the 2nd convolutional layer
Between the 3rd convolutional layer, between the 4th convolutional layer and the 5th convolutional layer, between the 6th convolutional layer and the 7th convolutional layer,
Between 8th convolutional layer and the 9th convolutional layer, between the 10th convolutional layer and the 11st convolutional layer, the 13rd convolutional layer and the 1st
Between a full articulamentum, it is all connected with 1 pond layer, and above-mentioned 13 convolutional layers and 3 full articulamentums use nonlinear activation
Function is handled, specifically can basis here to the number of plies of above-mentioned convolutional layer, full articulamentum and pond layer without limiting
Actual conditions are chosen, similarly, for activation primitive selected in every layer also without limiting.
204, when the processing for receiving sports video data is requested, by the sports video in sports video data to be processed
Picture is input to the action recognition model, obtains the technical movements in sports video picture.
In embodiments of the present invention, since action recognition model is the sports video for having technical movements label by preliminary making
Constructed by picture training, therefore, when the sports video picture that will be extracted from sports video data to be processed is input to movement
After identification model, weighted value of the sports video picture in different technologies movement can be exported, the higher explanation of weighted value is regarded from sport
In frequency picture identification obtain the technical movements probability it is higher, and then using the corresponding technical movements of weighted value highest as identifying
To the technical movements in sports video picture.
205, the sports video data of different angle and different distance based on acquisition equipment capture, regards the sport
Frequency obtains the sports video data of different acquisition angle and different acquisition distance according to being divided.
Under normal conditions, the sports video data of fixed angle and fixed range can be captured using fixing camera,
Certainly remote camera lens, medium shot and short distance camera lens be can be combined with and acquire obstructed over-angle and different distance respectively
Sports video data, here without limit.
It is understood that due to different angle and the collected sports video data of different distance to follow-up data at
Reason result has an impact, so screening more excellent video after sports video picture is marked for the ease of spectators and drawing
Face can more intuitively distinguish sports video data by dividing the sports video data of different distance and different angle.
It can identify the human body target in sport video pictures, by human testing algorithm specifically to determine that human body target exists
Coordinate and size in sports video picture specifically can use yolo target detection for each sports video picture, obtain
Human body target and human body coordinate and size into picture, the further coordinate according to human body target in sports video picture
And size, accounting value of the human body target in sports video picture is calculated, the length and width institute of target body can be specifically calculated
In threshold range, the accounting value based on human body target in sports video picture divides sports video data, obtains
The sports video data of different acquisition angle and different acquisition distance.
Under normal conditions, based on acquisition distance collected sports video data can be divided into it is remote, in, short distance camera lens
Collected sports video data can be divided into front, side and back side video data based on acquisition angles by video data.
206, based on the solicited message that carries in the processing request, to the technical movements in each sports video picture into
Row statistics, obtains the processing result of sports video data.
In embodiments of the present invention, when processing request is the frequency of occurrence of the first technical movements in statistics sports video data
When, carry the first technical movements of request statistics in the solicited message, and by obtaining to identification in sports video data
Technical movements carry out continuous sex determination, the frequency of occurrence of each technical movements in sports video data are counted, for example, sports video
Dunk shot technical movements occur 7 times in data, and secondary attack technical movements occur 10 times, and backboard technical movements appearance 4 is inferior, according to sport
The frequency of occurrence of each technical movements in video data, determines the frequency of occurrence of the first technical movements, obtains sports video data
Processing result.
For example, for the frequency of occurrence of a certain technical movements in statistics sports video data, since action recognition model can
To identify the technical movements in sports video data in each sports video picture, if there is continuous three or three with
The technical movements that upper sport video pictures identify are identical, then determine occur the primary technical movements in sports video data,
By there is the number of the technical movements in cumulative sports video data, a certain technical movements in sports video data can be obtained
Frequency of occurrence.
In embodiments of the present invention, when processing request is screens the wonderful in sports video data, request letter
Corresponding second technical movements of wonderful are carried in breath, and are based on the corresponding acquisition angles of sports video data and acquisition
Distance filters out acquisition angles and is positive and acquires the sports video data of distance within a preset range as excellent candidate
Section counts and goes out in candidate wonderful by carrying out continuous sex determination to the technical movements identified in candidate wonderful
Existing technical movements determine that the candidate for the second technical movements occur is excellent according to the technical movements occurred in candidate wonderful
Segment obtains the processing result of sports video data.
For example, preferentially picking out acquisition from sports video data for the wonderful in screening sports video data
Angle is front and acquires sports video data of the distance within the scope of 100 meters as candidate wonderful, similarly, due to dynamic
Technical movements in candidate wonderful in each sports video picture can be identified by making identification model, if there is continuous three
The technical movements that a or three or more sports video picture recognitions obtain are identical, then determine occur the skill in candidate wonderful
Art movement, so that the candidate wonderful for the technical movements occur to be determined as to the wonderful in sports video data.
It should be noted that being input to the recognition result that identification model obtains using sports video data is sports video figure
Probability scenarios of the piece in different technologies movement, are equivalent to the classification to technical movements, in classification output, if output be not
0-1, but real number value belong to the probability of each classification, then can be evaluated with confidence level recognition result, this is defeated
Probability indicates the confidence level of the movement of different technologies corresponding to sports video data out.For example, classifier output video data institute is right
The confidence level for answering dunk shot technical movements is 0.51, then illustrating that the classification results reliability is lower, then in screening sports video
During wonderful in data, preferentially the recognition result that confidence level is lower than default value can be deleted.
In addition, above-mentioned filtering out the sports video data of acquisition angles within a preset range for positive and acquisition distance
During as candidate wonderful, in order to guarantee the duration of wonderful, duration can also be carried out to candidate wonderful
Limitation, can candidate wonderful by candidate wonderful duration not within the scope of default value delete.
It is understood that the technical movements identified in sports video data or candidate wonderful carry out
It, specifically can be dynamic by the technology occurred in statistics sports video data or candidate wonderful during continuous sex determination
Make, based on the technical movements that each video pictures in sports video data or candidate wonderful identify, if sport
What the sports video picture recognition that continuously appearance is greater than or equal to preset quantity in video data or candidate wonderful obtained
Technical movements are identical, then determine occur a technical movements in sports video data or candidate wonderful, statistics sport view
The frequency of occurrence of each technical movements in frequency evidence or candidate wonderful.
Further, the specific implementation as Fig. 1 the method, the embodiment of the invention provides a kind of sports video datas
Processing unit, as shown in figure 3, described device includes: acquiring unit 31, construction unit 32, predicting unit 33, statistic unit
34。
Acquiring unit 31 can be used for obtaining sports video sample data, include to take in the sports video sample data
Sports video picture with technical movements label;
Construction unit 32 can be used for that the sport view of technical movements label will be carried in the sports video sample data
Frequency picture is input in deep learning model and is trained, and constructs action recognition model, the action recognition model is for identification
Technical movements in sports video picture;
Recognition unit 33 can be used for when the processing for receiving sports video data is requested, by sports video to be processed
Sports video picture in data is input to the action recognition model, obtains the technical movements in sports video picture;
Statistic unit 34 can be used for based on the solicited message carried in the processing request, to each sports video figure
Technical movements in piece are counted, and the processing result of sports video data is obtained.
The processing unit of a kind of sports video data provided by the invention, by obtaining sports video sample data, by body
It educates the sports video picture for carrying technical movements label that video sample data include and is input in deep learning model and carry out
Training constructs action recognition model, to be handled according to action recognition model sports video data, without artificial right
Sports video data is counted, and the technical movements occurred in sports video data can be automatically identified, and improves sports video
The processing accuracy of data.With in the prior art by the way of artificial statistics or artificial editing come to sports video data into
The mode of row processing is compared, and the present invention identifies the technical movements in sports video data by the technology identification model of building,
Without manual identified or artificial editing, the processing time of sports data is saved, since artificial treatment is there are certain subjectivity,
The processing result of sports video data based on the generation of technology identification model has higher precision, improves at sports video data
The flexibility of reason.
The further explanation of processing unit as sports video data shown in Fig. 3, Fig. 4 are according to embodiments of the present invention
The structural schematic diagram of the processing unit of another sports video data of offer, as shown in figure 4, described device further include:
Division unit 35 can be used for described when the processing for receiving sports video data is requested, by body to be processed
The sports video picture educated in video data is input to the action recognition model, obtains the technical movements in sports video picture
Later, the sports video data of different angle and different distance based on acquisition equipment capture, to the sports video data
It is divided, obtains the sports video data of different acquisition angle and different acquisition distance.
Further, the division unit 35 includes:
Identification module 351 can be used for identifying the human body target in sport video pictures by human testing algorithm, determine
Coordinate and size of the human body target in sports video picture;
Computing module 352 can be used for coordinate and size according to the human body target in sports video picture, meter
Calculate accounting value of the human body target in sports video picture;
Division module 353 can be used for the accounting value based on human body target in sports video picture and regard to the sport
Frequency obtains the sports video data of different acquisition angle and different acquisition distance according to being divided.
Further, the deep learning model is depth residual error network, and the depth residual error network includes multilayered structure,
The construction unit 32 includes:
Extraction module 321 can be used for extracting the spy of sports video picture by the convolutional layer of the depth residual error network
Sign obtains characteristic parameter of the sports video picture in different technologies movement;
Dimensionality reduction module 322 can be used for existing to the sports video picture by the pond layer of the depth residual error network
Characteristic parameter in different technologies movement carries out dimension-reduction treatment, and sports video picture is acted in different technologies after obtaining dimension-reduction treatment
On characteristic parameter;
Summarizing module 323, after can be used for summarizing the dimension-reduction treatment by the full articulamentum of the depth residual error network
Characteristic parameter of the sports video picture in different technologies movement obtains the power that different technologies movement is characterized in sports video picture
Weight values;
Generation module 324 can be used for the classification layer by the depth residual error network according to the sports video picture
The weighted value of middle characterization different technologies movement generates the recognition result of technical movements in sports video picture.
Further, the acquiring unit 31 includes:
Collection module 311 can be used for collecting the sports video data of acquisition equipment capture, according to prefixed time interval pair
The sports video data is split, and obtains multistage sports video frame data;
Mark module 312 can be used for educating from every segment body and choose multiple sports video pictures in video requency frame data, be based on body
The sports video picture is marked in the technical movements educated in video pictures, obtains sports video sample data.
Further, when the processing request is counts the frequency of occurrence of the first technical movements in sports video data,
The first technical movements of request statistics are carried in the solicited message, the statistic unit 34 includes:
First statistical module 341 can be used for by connecting to the technical movements identified in sports video data
Continuous sex determination, counts the frequency of occurrence of each technical movements in sports video data;
First determining module 342 can be used for the frequency of occurrence according to technical movements each in the sports video data,
The frequency of occurrence for determining the first technical movements obtains the processing result of sports video data;
First statistical module 341 specifically can be used for identifying based on each video pictures in sports video data
The technical movements arrived, if the sports video picture recognition for continuously occurring being greater than or equal to preset quantity in sports video data obtains
The technical movements arrived are identical, then determine occur a technical movements in sports video data, count each in sports video data
The frequency of occurrence of technical movements.
The further explanation of processing unit as sports video data shown in Fig. 3, Fig. 5 are according to embodiments of the present invention
The structural schematic diagram of the processing unit of another sports video data of offer, as shown in figure 5, when processing request is screening
When wonderful in sports video data, corresponding second technical movements of wonderful, institute are carried in the solicited message
Stating statistic unit 34 includes:
Screening module 343 can be used for based on the corresponding acquisition angles of sports video data and acquire distance, filters out
Acquisition angles are positive and acquire the sports video data of distance within a preset range as candidate wonderful;
Second statistical module 344 can be used for by connecting to the technical movements identified in candidate wonderful
Continuous sex determination counts the technical movements occurred in candidate wonderful;
Second determining module 345 can be used for determining and occurring according to the technical movements occurred in the candidate wonderful
The candidate wonderful of second technical movements, obtains the processing result of sports video data;
Second statistical module 344 specifically can be used for identifying based on each video pictures in candidate wonderful
The technical movements arrived, if the sports video picture recognition for continuously occurring being greater than or equal to preset quantity in candidate wonderful obtains
The technical movements arrived are identical, then determine occur a technical movements in candidate wonderful, count and occur in candidate wonderful
Technical movements.
It should be noted that each functional unit involved by a kind of processing unit of sports video data provided in this embodiment
Other it is corresponding describe, can be with reference to the corresponding description in Fig. 1 and Fig. 2, details are not described herein.
It is deposited thereon based on above-mentioned method as depicted in figs. 1 and 2 correspondingly, the present embodiment additionally provides a kind of storage medium
Computer program is contained, which realizes the processing of above-mentioned sports video data as depicted in figs. 1 and 2 when being executed by processor
Method.
Based on this understanding, the technical solution of the application can be embodied in the form of software products, which produces
Product can store in a non-volatile memory medium (can be CD-ROM, USB flash disk, mobile hard disk etc.), including some instructions
With so that computer equipment (can be personal computer, server or the network equipment an etc.) execution the application is each
Method described in implement scene.
Based on above-mentioned method as shown in Figure 1 and Figure 2 and Fig. 3, Fig. 4, virtual bench embodiment shown in fig. 5, in order to
Realize above-mentioned purpose, the embodiment of the present application also provides a kind of computer equipment, be specifically as follows personal computer, server,
Network equipment etc., the entity device include storage medium and processor;Storage medium, for storing computer program;Processor,
The processing method of above-mentioned sports video data as depicted in figs. 1 and 2 is realized for executing computer program.
Optionally, which can also include user interface, network interface, camera, radio frequency (Radio
Frequency, RF) circuit, sensor, voicefrequency circuit, WI-FI module etc..User interface may include display screen
(Display), input unit such as keyboard (Keyboard) etc., optional user interface can also connect including USB interface, card reader
Mouthful etc..Network interface optionally may include standard wireline interface and wireless interface (such as blue tooth interface, WI-FI interface).
It will be understood by those skilled in the art that the entity device of the processing unit of sports video data provided in this embodiment
Structure does not constitute the restriction to the entity device, may include more or fewer components, perhaps combine certain components or
Different component layouts.
It can also include operating system, network communication module in storage medium.Operating system is that the above-mentioned computer of management is set
The program of standby hardware and software resource, supports the operation of message handling program and other softwares and/or program.Network communication mould
Block leads to for realizing the communication between each component in storage medium inside, and between other hardware and softwares in the entity device
Letter.
Through the above description of the embodiments, those skilled in the art can be understood that the application can borrow
It helps software that the mode of necessary general hardware platform is added to realize, hardware realization can also be passed through.Pass through the skill of application the application
Art scheme, compared with currently available technology, the present invention is identified in sports video data by the technology identification model of building
Technical movements are not necessarily to manual identified or artificial editing, save the processing time of sports data, since artificial treatment exists centainly
Subjectivity, based on technology identification model generate sports video data processing result have higher precision, improve sport
The flexibility of video data processing.
It will be appreciated by those skilled in the art that the accompanying drawings are only schematic diagrams of a preferred implementation scenario, module in attached drawing or
Process is not necessarily implemented necessary to the application.It will be appreciated by those skilled in the art that the mould in device in implement scene
Block can according to implement scene describe be distributed in the device of implement scene, can also carry out corresponding change be located at be different from
In one or more devices of this implement scene.The module of above-mentioned implement scene can be merged into a module, can also be into one
Step splits into multiple submodule.
Above-mentioned the application serial number is for illustration only, does not represent the superiority and inferiority of implement scene.Disclosed above is only the application
Several specific implementation scenes, still, the application is not limited to this, and the changes that any person skilled in the art can think of is all
The protection scope of the application should be fallen into.
Claims (10)
1. a kind of processing method of sports video data, which is characterized in that the described method includes:
Sports video sample data is obtained, is regarded in the sports video sample data comprising carrying the sport of technical movements label
Frequency picture;
The sports video picture that technical movements label is carried in the sports video sample data is input to deep learning mould
It is trained in type, constructs action recognition model, the technology in sports video picture is dynamic for identification for the action recognition model
Make;
When the processing for receiving sports video data is requested, the sports video picture in sports video data to be processed is inputted
To the action recognition model, the technical movements in sports video picture are obtained;
Based on the solicited message carried in the processing request, the technical movements in each sports video picture are counted,
Obtain the processing result of sports video data.
2. the method according to claim 1, wherein the deep learning model be depth residual error network, it is described
Depth residual error network includes multilayered structure, the sport that technical movements label will be carried in the sports video sample data
Video pictures are input in deep learning model and are trained, and building action recognition model includes:
The feature that sports video picture is extracted by the convolutional layer of the depth residual error network, obtains sports video picture in difference
Characteristic parameter in technical movements;
Characteristic parameter by the pond layer of the depth residual error network to the sports video picture in different technologies movement
Dimension-reduction treatment is carried out, characteristic parameter of the sports video picture in different technologies movement after dimension-reduction treatment is obtained;
Sports video picture is dynamic in different technologies after summarizing the dimension-reduction treatment by the full articulamentum of the depth residual error network
Characteristic parameter on work obtains the weighted value that different technologies movement is characterized in sports video picture;
By the classification layer of the depth residual error network according to the weight for characterizing different technologies movement in the sports video picture
Value generates the recognition result of technical movements in sports video picture.
3. the method according to claim 1, wherein the acquisition sports video sample data, specifically includes:
The sports video data for collecting acquisition equipment capture, divides the sports video data according to prefixed time interval
It cuts, obtains multistage sports video frame data;
It is educated from every segment body and chooses multiple sports video pictures in video requency frame data, based on the technical movements pair in sports video picture
The sports video picture is marked, and obtains sports video sample data.
4. the method according to any one of right 1, which is characterized in that when processing request is statistics sports video data
In the first technical movements frequency of occurrence when, the first technical movements of request statistics, the base are carried in the solicited message
The solicited message carried in the processing request, counts the technical movements in each sports video picture, obtains body
The processing result for educating video data includes:
By carrying out continuous sex determination to the technical movements identified in sports video data, count every in sports video data
The frequency of occurrence of a technical movements;
According to the frequency of occurrence of technical movements each in the sports video data, the frequency of occurrence of the first technical movements is determined,
Obtain the processing result of sports video data;
It is described by carrying out continuous sex determination to the technical movements that identify in sports video data, count sports video data
In each technical movements frequency of occurrence, specifically include:
Based on the technical movements that video pictures each in sports video data identify, if continuously gone out in sports video data
It is now identical more than or equal to the technical movements that the sports video picture recognition of preset quantity obtains, then determine in sports video data
There are a technical movements, counts the frequency of occurrence of each technical movements in sports video data.
5. the method according to claim 1, wherein described when the processing request for receiving sports video data
When, the sports video picture in sports video data to be processed is input to the action recognition model, obtains sports video figure
After technical movements in piece, the method also includes:
Based on acquisition equipment capture different angle and different distance sports video data, to the sports video data into
Row divides, and obtains the sports video data of different acquisition angle and different acquisition distance.
6. according to the method described in claim 5, it is characterized in that, the different angle and not based on acquisition equipment capture
The sports video data of same distance divides the sports video data, specifically includes:
The human body target in sport video pictures is identified by human testing algorithm, determines human body target in sports video picture
Coordinate and size;
According to coordinate and size of the human body target in sports video picture, human body target is calculated in sports video picture
In accounting value;
Accounting value based on human body target in sports video picture divides the sports video data, obtains difference and adopts
Collect the sports video data of angle and different acquisition distance.
7. the method according to any one of right 1-6, which is characterized in that when processing request is screening sport video counts
When wonderful in, corresponding second technical movements of wonderful are carried in the solicited message, it is described based on described
The solicited message carried in processing request, counts the technical movements in each sports video picture, obtains sports video
The processing result of data includes:
Based on the corresponding acquisition angles of sports video data and acquisition distance, filter out acquisition angles for positive and acquisition away from
Sports video data within a preset range is as candidate wonderful;
By carrying out continuous sex determination to the technical movements identified in candidate wonderful, counts and go out in candidate wonderful
Existing technical movements;
According to the technical movements occurred in the candidate wonderful, determine the candidate wonderful of the second technical movements occur,
Obtain the processing result of sports video data;
It is described by carrying out continuous sex determination to the technical movements that identify in candidate wonderful, count candidate wonderful
The technical movements of middle appearance, specifically include:
Based on the technical movements that each video pictures in candidate wonderful identify, if continuously gone out in candidate wonderful
It is now identical more than or equal to the technical movements that the sports video picture recognition of preset quantity obtains, then determine in candidate wonderful
There are a technical movements, counts the technical movements occurred in candidate wonderful.
8. a kind of processing unit of sports video data, which is characterized in that described device includes:
Acquiring unit is moved in the sports video sample data comprising carrying technology for obtaining sports video sample data
Make the sports video picture of label;
Construction unit, for inputting the sports video picture for carrying technical movements label in the sports video sample data
It is trained into deep learning model, constructs action recognition model, action recognition model sports video figure for identification
Technical movements in piece;
Recognition unit, for when receive sports video data processing request when, by the body in sports video data to be processed
It educates video pictures and is input to the action recognition model, obtain the technical movements in sports video picture;
Statistic unit, the solicited message for being carried in being requested based on the processing, to the technology in each sports video picture
Movement is counted, and the processing result of sports video data is obtained.
9. a kind of computer equipment, including memory and processor, the memory are stored with computer program, feature exists
In the step of processor realizes any one of claims 1 to 7 the method when executing the computer program.
10. a kind of computer storage medium, is stored thereon with computer program, which is characterized in that the computer program is located
The step of reason device realizes method described in any one of claims 1 to 7 when executing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910624923.7A CN110490064B (en) | 2019-07-11 | 2019-07-11 | Sports video data processing method and device, computer equipment and computer storage medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910624923.7A CN110490064B (en) | 2019-07-11 | 2019-07-11 | Sports video data processing method and device, computer equipment and computer storage medium |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110490064A true CN110490064A (en) | 2019-11-22 |
CN110490064B CN110490064B (en) | 2024-05-14 |
Family
ID=68547005
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910624923.7A Active CN110490064B (en) | 2019-07-11 | 2019-07-11 | Sports video data processing method and device, computer equipment and computer storage medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110490064B (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114245206A (en) * | 2022-02-23 | 2022-03-25 | 阿里巴巴达摩院(杭州)科技有限公司 | Video processing method and device |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9600717B1 (en) * | 2016-02-25 | 2017-03-21 | Zepp Labs, Inc. | Real-time single-view action recognition based on key pose analysis for sports videos |
CN109886165A (en) * | 2019-01-23 | 2019-06-14 | 中国科学院重庆绿色智能技术研究院 | A kind of action video extraction and classification method based on moving object detection |
-
2019
- 2019-07-11 CN CN201910624923.7A patent/CN110490064B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9600717B1 (en) * | 2016-02-25 | 2017-03-21 | Zepp Labs, Inc. | Real-time single-view action recognition based on key pose analysis for sports videos |
CN109886165A (en) * | 2019-01-23 | 2019-06-14 | 中国科学院重庆绿色智能技术研究院 | A kind of action video extraction and classification method based on moving object detection |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114245206A (en) * | 2022-02-23 | 2022-03-25 | 阿里巴巴达摩院(杭州)科技有限公司 | Video processing method and device |
CN114245206B (en) * | 2022-02-23 | 2022-07-15 | 阿里巴巴达摩院(杭州)科技有限公司 | Video processing method and device |
Also Published As
Publication number | Publication date |
---|---|
CN110490064B (en) | 2024-05-14 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2020151489A1 (en) | Living body detection method based on facial recognition, and electronic device and storage medium | |
US10965886B2 (en) | System and method of generating a composite frame | |
Tran et al. | Two-stream flow-guided convolutional attention networks for action recognition | |
CN106028134A (en) | Detect sports video highlights for mobile computing devices | |
CN109376603A (en) | A kind of video frequency identifying method, device, computer equipment and storage medium | |
CN108810620A (en) | Identify method, computer equipment and the storage medium of the material time point in video | |
US20150092981A1 (en) | Apparatus and method for providing activity recognition based application service | |
CN109684919B (en) | Badminton service violation distinguishing method based on machine vision | |
US11568617B2 (en) | Full body virtual reality utilizing computer vision from a single camera and associated systems and methods | |
CN111291617B (en) | Badminton event video highlight segment extraction method based on machine learning | |
CN108905095B (en) | Athlete competition state evaluation method and equipment | |
CN112036209A (en) | Portrait photo processing method and terminal | |
CN113111690B (en) | Facial expression analysis method and system and satisfaction analysis method and system | |
CN114863321B (en) | Automatic video generation method and device, electronic equipment and chip system | |
Mehrasa et al. | Learning person trajectory representations for team activity analysis | |
CN110490064A (en) | Processing method, device, computer equipment and the computer storage medium of sports video data | |
CN113297883A (en) | Information processing method, analysis model obtaining device and electronic equipment | |
CN116524393A (en) | Badminton analysis system and badminton analysis method | |
CN115475373A (en) | Motion data display method and device, storage medium and electronic device | |
Jiang et al. | Deep learning application in broadcast tennis video annotation | |
WO2021078037A1 (en) | Facial skin detection method and apparatus | |
Stefański et al. | The problem of detecting boxers in the boxing ring | |
CN113642531A (en) | Video image recognition method, system and computer readable storage medium | |
CN113876311B (en) | Non-contact type multi-player heart rate efficient extraction device capable of adaptively selecting | |
Gautam et al. | An Overview of Human Activity Recognition from Recordings |
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 |