CN113268515A - Automatic explanation device and method for football match - Google Patents
Automatic explanation device and method for football match Download PDFInfo
- Publication number
- CN113268515A CN113268515A CN202110597834.5A CN202110597834A CN113268515A CN 113268515 A CN113268515 A CN 113268515A CN 202110597834 A CN202110597834 A CN 202110597834A CN 113268515 A CN113268515 A CN 113268515A
- Authority
- CN
- China
- Prior art keywords
- football
- event
- text
- events
- team
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 29
- 238000001514 detection method Methods 0.000 claims abstract description 20
- 230000015572 biosynthetic process Effects 0.000 claims abstract description 18
- 238000003786 synthesis reaction Methods 0.000 claims abstract description 18
- 238000013145 classification model Methods 0.000 claims description 11
- 230000007704 transition Effects 0.000 claims description 9
- 230000008030 elimination Effects 0.000 claims description 5
- 238000003379 elimination reaction Methods 0.000 claims description 5
- 238000002372 labelling Methods 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 230000006735 deficit Effects 0.000 claims 1
- 238000005516 engineering process Methods 0.000 abstract description 11
- 238000010801 machine learning Methods 0.000 abstract description 3
- 238000003058 natural language processing Methods 0.000 abstract 1
- 238000010586 diagram Methods 0.000 description 4
- 102100033814 Alanine aminotransferase 2 Human genes 0.000 description 2
- 101000779415 Homo sapiens Alanine aminotransferase 2 Proteins 0.000 description 2
- 230000006378 damage Effects 0.000 description 2
- 238000005406 washing Methods 0.000 description 2
- 238000004140 cleaning Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/245—Query processing
- G06F16/2457—Query processing with adaptation to user needs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/24—Querying
- G06F16/248—Presentation of query results
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Molecular Biology (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Databases & Information Systems (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention relates to an automatic explanation device and method for a football match, and belongs to the technical field of machine learning. The method comprises the steps of firstly, constructing a player and team database containing player and team basic information and recent news; then, current match condition information such as event types, players corresponding to the events, event occurrence positions, event occurrence time and the like in the football match is obtained by utilizing a video event detection technology and a video tracking technology; selectively outputting football commentary of different categories such as player team introduction, player team news abstract, game event commentary, ending word and the like according to the game situation, wherein the game event commentary is generated by a text generation technology according to the game condition information; finally, the commentary text is converted into corresponding audio by using a speech synthesis technology. The method realizes the automatic explanation of the football game through the technologies of natural language processing, computer vision, voice synthesis and the like, saves the labor cost for the commentator and enables fans to enjoy more games with explanation.
Description
Technical Field
The invention relates to the field of sports event explanation, in particular to an automatic explanation device and method for a football game, and belongs to the technical field of machine learning.
Technical Field
Football attracts thousands of fans as the first sport in the world. Fans often need professional commentary to enhance their viewing experience while watching the game. However, due to the overhead of labor cost and the scarcity of excellent commentators, in some low-level tournaments, there is often only a video source and no commentator, which greatly reduces the watching pleasure of fans.
However, with the rise of deep learning, the video analysis, text generation and voice synthesis technology realizes great progress and creates conditions for the realization of the automatic explanation system of the football match. The automatic explanation system for the football match can explain the scene of the match, background information of teams of players and news of near conditions in real time, and is very similar to the real explanation, so that the problems of small number of commentators and high labor cost are well solved, and audiences enjoy professional explanation in more matches.
Disclosure of Invention
The invention aims to solve the technical problem of how to realize automatic explanation of a football game by machine learning, and creatively provides a device and a method for automatically explaining the football game. A player team database is first constructed containing basic information about players and teams, as well as recent news. And then, current match condition information such as a football event, a player corresponding to the event, an event occurrence position, event occurrence time and the like in the football match is obtained by utilizing a video event detection technology and a video tracking technology. And then selectively outputting different categories of football commentary such as player team introduction, player team news abstract, game event commentary, ending words and the like according to the game situation, wherein the game event commentary is generated by a text generation technology according to the game condition information. Finally, the commentary text is converted into corresponding audio by using a speech synthesis technology.
Specifically, the technical scheme adopted by the invention is as follows:
an automatic commentary method for a football match comprises the following steps:
s10, establishing a football team database of the football match, and establishing a football team text information acquisition system;
s20, completing match event identification and player matching according to the video information of the football match, and constructing a video information acquisition system; the events include: pass, take a ball, shoot, goal, corner ball, free kick, point kick, out of bounds, goal ball, salvage, free from bounds, put out of bounds, go out of sideline, go out of bottom line, foul, red card, yellow card;
s30, generating commentary text according to the game event and the related information, and constructing a commentary generation system; the explanation text comprises player team introduction, player team news abstract text, match event explanation text and end words;
s40, generating the comment audio from the output comment text by using a speech synthesis model, and constructing a speech synthesis system.
Further, the player team database includes:
the player and team structural information comprises player basic information, player historical information, team basic information and team historical information; wherein, sportsman's basic information is: name, age, height, weight, position of employment, and jersey number; the historical information of the player is as follows: effectiveness team, start and stop time, conference fee, historical goal number in each game, attack number, time of getting on the market, and honor gained; the basic information of the team is as follows: team name, city where team is located, and main court name; the team history information is: the opponent team and the score of each historical match, the tournament name and the rank of each historical season, and the honor obtained by the team;
text categories of player and/or team news: taking news of players and/or teams in a period D and corresponding browsing amount as input, and outputting text categories of the players and/or teams according to a trained text classification model, such as a Bert model; the text classification model is obtained by taking the collected M pieces of football news as a data set of the text classification model, dividing the football news into five types of after-match reports, event foregoers, player injuries, meeting news and other news, training the model until the model converges, and storing the converged text classification model parameters;
text summary of player and/or team news and its categorization: taking news of the players and/or teams in a period D and corresponding browsing amount as input, generating a model such as a GPT2 model according to the trained text abstract, outputting text abstract of the players and/or teams, and classifying the text abstract of the players and/or teams according to the text category; the text abstract generating model is obtained by training a model until the model converges according to the collected N football news serving as a data set of the text abstract generating model and the text abstract corresponding to the football news serving as the output of the model, and storing the converged text abstract generating model parameters.
Further, the match event identification and player matching comprises:
s201, collecting a plurality of football match videos, segmenting the football match videos into a plurality of short videos containing single match events, and labeling event types to form a football match event data set;
s202, constructing an event detection model, training on the football event data set by using a video event scene recognition algorithm, such as a NetVLAD algorithm, and a background elimination algorithm, and storing model parameters after the model converges;
s203, constructing a football game event state machine, setting a certain time length t as the time length of a detection window, inputting a game video, obtaining the type of the game event output by the current detection window by combining the transition probability of the state machine and the event identification probability of the event detection model, and simultaneously recording the time information of the detection window; the transition probability of the state machine is shown in table 1, and each numerical value in the table represents the probability of the event corresponding to the column where the numerical value is located transitioning to the event corresponding to the row where the numerical value is located;
s204, detecting the track of the football by using a video detection model, such as a YOLO4 model, and a background elimination algorithm to obtain the position and time information of the football; detecting the tracks of the players by using a video tracking algorithm, such as a deepsort algorithm, and obtaining player position information and moment information;
s205, matching the match event at the current moment with the player of the event according to the match event category and the time information obtained in S203, the player position information and the time information and the football position and the time information obtained in S204; and obtaining current match condition information consisting of event types, players corresponding to the events, event occurrence positions and event occurrence time.
TABLE 1 event State transition probability Table
Further, the commentary generation system judges the competition state according to the competition event type, the event occurrence position and the time information, wherein the competition state comprises opening, finishing, tension and relaxation. If the match state is open, outputting player team introduction in the database in the player team text information acquisition system; if the competition state is the end, outputting an end word consisting of the names of the two teams and the competition score; if the competition state is judged to be tension, inputting a triple composed of the event type, players corresponding to the event and the event occurrence position into a text generation model according to a pre-trained football comment text generation model, such as an LSTM model, and outputting a comment text of the competition event; and if the match state is judged to be moderate, outputting the news abstract text of the football teams in the database in the football team text information acquisition system.
Still further, the method for generating the football comment text generation model comprises the following steps: and training a football comment text generation model according to a plurality of collected and sorted football game comment texts serving as data sets of the football comment text generation model, mapping a triple formed by the input event type, players corresponding to the event and the event occurrence position into a comment text, and storing model parameters after the model converges.
Further, the speech synthesis system converts the game event commentary text, or the player team introduction, or the player team news abstract text, or the end word into corresponding audio and outputs the audio by using a trained speech synthesis model, such as a Tacotron2 model.
Meanwhile, the invention also discloses an automatic explanation device for the football match, which comprises:
the football team text information acquisition system is used for establishing a football team database of a football match;
the video information acquisition system is used for completing match event identification and player matching according to the video information of the football match; the events include: pass, take a ball, shoot, goal, corner ball, free kick, point kick, out of bounds, goal ball, salvage, free from bounds, put out of bounds, go out of sideline, go out of bottom line, foul, red card, yellow card;
the commentary generation system is used for generating commentary texts according to the game events and the related information; the explanation text comprises player team introduction, player team news abstract text, match event explanation text and end words;
and the speech synthesis system generates the comment audio by using a speech synthesis model to the output comment text.
Further, the automatic narration device executes the steps of the method. Particularly, the video information acquisition system comprises a match event identification module, a player track tracking module and a football track tracking module; the match event recognition module performs the steps of S202 and S203, and the player trajectory tracking module and the football trajectory tracking module perform the step of S204.
Drawings
Fig. 1 is a flow chart of an automatic explanation method of a soccer game.
Fig. 2 is a schematic diagram of an automatic explanation device for a football game.
Fig. 3 is a competition state decision diagram.
Detailed Description
The invention is further illustrated and described in detail below with reference to the figures and examples.
Fig. 1 is a schematic flow chart of an embodiment of the method of the present invention, fig. 2 is a schematic diagram of an automatic explanation device for a soccer game, and fig. 3 is a game state decision diagram. The automatic explanation method for the football match comprises the following specific steps:
step 1: and constructing a player team text information acquisition system and establishing a player team database. The method comprises the following steps:
step 1.1: structured information of players and teams is collected by applying a crawler technology, and the information is stored in a database after being cleaned and arranged. Including player basic information: name, age, height, weight, position of employment, jersey number, etc.; history information of the player: effectiveness team, start and stop time, meeting change fee and the like, the number of goals, attack-assisting numbers, time of getting on the scene and the like in each historical game, and the honor of players; team basic information: team name, team city, main court name, etc.; team history information: the opponent team and the score of each game in history, the tournament name and the ranking of each season in history and the honor gained by the team.
Step 1.2: a large number of football news are collected as a data set of a text classification model and a text abstract generation model. And performing data cleaning on the acquired news text, including removing information (such as news authors, special characters and the like) irrelevant to the news content.
Step 1.3: and (3) constructing a text classification model (such as Bert), classifying the news of the players and the teams obtained in the step 1.2 into five categories of post-event reports, event forelookers, player injuries, meeting news and other news, and training the model. And saving the parameters after the model converges. An L2 regularization and dropout method are introduced during training to mitigate model overfitting. In addition, to reduce redundant information, only the first 128 characters of news and the last 382 characters of news are input into the model.
Step 1.4: and (3) constructing a text abstract generating model (such as GPT2), taking the football news obtained in the step 1.2 and the corresponding text abstract as the input and the output of the model, and training the model. And saving the parameters after the model converges. An L2 regularization and dropout method are introduced during training to mitigate model overfitting. In addition, to ensure that the summary is not too long, the output of the model is limited to 45 characters.
Step 1.5: and collecting recent (within one week) news and corresponding browsing amount of the corresponding player and team according to the name of the player and the name of the team.
Step 1.6: and (3) after data washing is carried out on the news of the players or teams obtained in the step 1.5, inputting the text classification model trained in the step 1.3, and outputting the text type of the model.
Step 1.7: and (3) after data washing is carried out on the news of the players or teams obtained in the step 1.5, inputting the text abstract trained in the step 1.4 to generate a model, outputting the text abstract, and classifying the text abstract according to the text category obtained in the step 1.6. And sorting according to the news browsing amount, and selecting 5 top-ranked text abstracts of each player or team under each category and storing the abstracts into a database.
Step 1.8: steps 1.5 to 1.7 are performed regularly each day, so that the text summaries of the players or teams stored in the database are always news in one week.
Step 2: and constructing a video information acquisition system to finish match event identification and player matching. The method comprises the following steps:
step 2.1: defining events in the football match, collecting a plurality of football match videos, segmenting the football match videos into a plurality of short videos containing single events, and labeling event categories to form a football match event data set. The events include: pass, take a ball, shoot, goal, corner ball, free kick, point kick, ball out of bounds, goal ball, salvage, free from bounds, put out of bounds, go out of sideline, go out of bottom line, foul, red card, yellow card.
Step 2.2: firstly, acquiring key frames in video clips by using an attention mechanism, training on the football event data set established in the step 2.1 by using a video event scene recognition algorithm (such as NetVLAD), and storing model parameters after the model converges. An L2 regularization and dropout method are introduced during training to mitigate model overfitting.
Step 2.3: and constructing a football game event state machine, setting 10s as the duration of a detection window, combining the transition probability of the state machine and the event identification probability of an event detection model to obtain the game event output by the current detection window, and simultaneously recording the time information of the detection window. The transition probabilities between events in the football game event state machine are shown in table 1, and each value represents the probability of the corresponding event in the column being transitioned to the corresponding event in the row being reached.
Step 2.4: the track of the football is tracked by using a video detection model (such as YOLO4) and a background elimination algorithm, and the track of the football is tracked by adopting a tracker connected in parallel. The method comprises the steps of detecting tracks of players by using a multi-target tracking algorithm (such as deepsort), calibrating the players in a rectangular frame by using a video detection algorithm during specific implementation, and extracting features in the rectangular frame by using a trained feature extractor (such as CNN), so that further feature matching is provided for the tracking algorithm, and the tracking robustness is improved.
Step 2.5: matching the match event at the current moment with the player of the event according to the match event category, the player position information and the moment information, and the football position and the moment information obtained in the steps 2.3 and 2.4. And obtaining four-dimensional information of event type, time corresponding to the players, event occurrence position and event occurrence time. The event occurrence positions are defined as a back field (1/3 behind the court with the attack direction as a reference), a middle field (1/3 to 2/3 behind the court with the attack direction as a reference), a front field (1/3 in front of the court and in a non-forbidden zone with the attack direction as a reference), and four areas in a forbidden zone.
And step 3: and constructing a commentary generation system and outputting an commentary text. The method comprises the following steps:
step 3.1: and (4) judging the competition state according to the event occurrence position and the time information obtained in the step 2.5. The specific decision flow is shown in fig. 3, and the game is divided into three time periods, i.e., 5 minutes before the start of the game, 5 minutes after the start of the game until the end of the game, and the end of the game, according to the time. Then, the four-analogy race states of 'opening the field', 'finishing the field', 'tension', 'relaxation' are output according to the position of the event in the race. Meanwhile, when the event section of the game is from 5 minutes to the end of the game and the event occurrence position is the back field or the middle field, two states of 'moderate' or 'tense' need to be randomly output according to probability.
Step 3.2: 19216 football game comment texts are collected and sorted and serve as data sets of the football game comment text generation model.
Step 3.3: training a football comment text generation model (such as LSTM), and mapping a triple formed by the input event type, the player corresponding to the event and the event occurrence position into a comment text. To ensure that the generated text is not too long, the output of the model is limited to within 30 characters.
Step 3.4: and if the competition state output in the step 3.1 is 'tension', inputting the triple consisting of the event type, the time corresponding to the player and the event occurrence position obtained in the step 2.5 into the text generation model trained in the step 3.3, and outputting an explanation text of the competition event.
Step 3.5: if the match state output in the step 3.1 is 'moderate', outputting the news abstract text of the football team of the player obtained in the step 1.7; if the football game is 'on the scene', outputting the player team introduction obtained in the step 1.1; if the result is 'end of the field', outputting an end word consisting of two team names and a game score (for example, 'the game is ended, Guangzhou Hengda 1: 1 war Hebeijing national security').
And 4, step 4: and constructing a voice synthesis system and generating the comment audio. The game event commentary text generated in step 3.4 or the player (team) introduction or news digest text obtained in step 3.5 is converted into corresponding audio and output using a trained speech synthesis model (e.g., tacontron 2).
Claims (10)
1. An automatic commentary method for a football match is characterized by comprising the following steps:
s10, establishing a football team database of the football match, and establishing a football team text information acquisition system;
s20, completing match event identification and player matching according to the video information of the football match, and constructing a video information acquisition system; the events include: pass, take a ball, shoot, goal, corner ball, free kick, point kick, out of bounds, goal ball, salvage, free from bounds, put out of bounds, go out of sideline, go out of bottom line, foul, red card, yellow card;
s30, generating commentary text according to the game event and the related information, and constructing a commentary generation system; the explanation text comprises player team introduction, player team news abstract text, match event explanation text and end words;
s40, generating the comment audio from the output comment text by using a speech synthesis model, and constructing a speech synthesis system.
2. An automatic commentary method for football match as claimed in claim 1, wherein said player team database comprises: the player and team structural information comprises player basic information, player historical information, team basic information and team historical information; wherein, sportsman's basic information is: name, age, height, weight, position of employment, and jersey number; the historical information of the player is as follows: effectiveness team, start and stop time, conference fee, historical goal number in each game, attack number, time of getting on the market, and honor gained; the basic information of the team is as follows: team name, city where team is located, and main court name; the team history information is: the opponent team and the score of each historical match, the tournament name and the rank of each historical season, and the honor obtained by the team;
text categories of player and/or team news: taking news of the players and/or teams and corresponding browsing amount in a period D as input, and outputting text categories of the players and/or teams according to the trained text classification model;
text summary of player and/or team news and its categorization: and taking news and corresponding browsing amount of the players and/or the teams in a period D as input, generating a model according to the trained text abstract, outputting the text abstract of the players and/or the teams, and classifying the text abstract of the players and/or the teams according to the text category.
3. An automatic commentary method for football match as claimed in claim 2, wherein said text classification model is obtained by taking the collected M pieces of football news as the data set of the text classification model, dividing the football news into five categories of post-match reports, advance-looking events, impairment, meeting news and other news, training the model until the model converges, and storing the converged text classification model parameters; the text abstract generating model is obtained by taking the collected N football news as a data set of the text abstract generating model, taking the text abstract corresponding to the football news as the output of the model, training the model until the model converges, and storing the converged text abstract generating model parameters.
4. A method for automatically commenting on a soccer game according to claim 1, wherein said match event recognition and player matching comprises:
s201, collecting a plurality of football match videos, segmenting the football match videos into a plurality of short videos containing single match events, and labeling event types to form a football match event data set;
s202, constructing an event detection model, training on the football event data set by using a video event scene recognition algorithm and a background elimination algorithm, and storing model parameters after the model converges;
s203, constructing a football game event state machine, setting a certain time length t as the time length of a detection window, inputting a game video, obtaining the type of the game event output by the current detection window by combining the transition probability of the state machine and the event identification probability of the event detection model, and simultaneously recording the time information of the detection window;
s204, detecting the track of the football by using a video detection model and a background elimination algorithm to obtain the position and time information of the football; detecting the tracks of the players by using a video tracking algorithm to obtain position information and time information of the players;
s205, matching the match event at the current moment with the player of the event according to the match event category and the time information obtained in S203, the player position information and the time information and the football position and the time information obtained in S204; and obtaining current match condition information consisting of event types, players corresponding to the events, event occurrence positions and event occurrence time.
5. An automatic commentary method for football match, as claimed in claim 4, characterized in that the transition probability of said state machine means: the pass may be transferred to one of the events of pass, carry, shoot, break, free, put out a fire, go out a sideline, go out a bottom line, foul and offside, and the probability of transferring to other events is 0; the ball may be transferred to one of the events of passing, dribbling, shooting, breaking, sideline outgoing, bottom line outgoing and foul, and the probability of transferring to other events is 0; the shooting may be transferred to one of the events of shooting, goal, encirclement, saving, sideline outgoing, bottom line outgoing and foul, and the probability of transferring to other events is 0; the probability of the goal transition to the goal is 1, and the probability of the transition to other events is 0; the probability that the corner ball is possibly transferred into one of the events of passing, carrying the ball, shooting, getting around, putting out a rescue, going out a sideline, going out a bottom line, doing a foul and changing a person is 0; any ball may be transferred into one of the events of passing, carrying, shooting, entering, getting around, putting out a rescue, going out a sideline, going out a bottom line, breaking a rule, offside and changing a person, and the probability of transferring into other events is 0; the probability that the point ball is possibly transferred to one of the events of goal, putting out a fire and getting out of the bottom line and is transferred to other events is 0; the out-of-bounds ball may be transferred to one of the events of passing, carrying, shooting, breaking, unbounding and foul, and the probability of transferring to the other event is 0; the goal ball may be transferred into one of the events of passing, carrying, shooting, cutting, getting off, and breaking rules, and the probability of transferring into other events is 0; the probability that the robbery break is possibly transferred into one of the events of passing, carrying a ball, shooting, going out a sideline, going out a bottom line and breaking a rule is 0; the unbounding may be transferred to one of the events of passing, carrying, shooting, sidelining, bottom line leaving and foul, and the probability of transferring to other events is 0; the first aid may be transferred to one of the events of passing, carrying, shooting, getting around, going out of sideline, going out of bottom line and foul, and the probability of transferring to other events is 0; the outgoing line is transferred into one of events of an outlier and a person-changing event, and the probability of transferring into other events is 0; the bottom line may be transferred to one of the events of corner ball, goal ball and person changing, and the probability of transferring to other events is 0; the foul may be transferred to one of the events of free kick, play and change, and the probability of transferring to other events is 0; the card-playing may be transferred to one of the events of free kick, click, card-playing and person-changing, and the probability of transferring to other events is 0; the offside can be transferred into one of the events of the free ball and the person changing, and the probability of transferring into other events is 0; the probability of the opening ball to be transferred to the ball carrying is 1, and the probability of the transfer to other events is 0; the person changing may be transferred to one of the events of corner ball, free kick, nod, out-of-bounds kick, goal kick and person changing, with a probability of 0 for transferring to the other event.
6. An automatic commentary method for football match as claimed in claim 1, wherein said commentary generation system determines the match state including opening, finishing, tension, and relaxation according to the match event category, the event occurrence position and the time information; if the match state is open, outputting player team introduction in the database in the player team text information acquisition system; if the competition state is the end, outputting an end word consisting of the names of the two teams and the competition score; if the competition state is judged to be tension, generating a model according to a pre-trained football comment text, inputting a triple composed of the event type, players corresponding to the event and the event occurrence position into the text generation model, and outputting a comment text of the competition event; and if the match state is judged to be moderate, outputting the news abstract text of the football teams in the database in the football team text information acquisition system.
7. An automatic commentary method for football match according to claim 6, wherein the football commentary text generation model is generated by: and training a football comment text generation model according to a plurality of collected and sorted football game comment texts serving as data sets of the football comment text generation model, mapping a triple formed by the input event type, players corresponding to the event and the event occurrence position into a comment text, and storing model parameters after the model converges.
8. An automatic commentary method for football match as claimed in claim 1, wherein said speech synthesis system uses trained speech synthesis model to convert said commentary text of match event, or football team introduction, or football team news abstract text, or end word into corresponding audio and output.
9. An automatic commentary device of a football match, characterized by comprising:
the football team text information acquisition system is used for establishing a football team database of a football match;
the video information acquisition system is used for completing match event identification and player matching according to the video information of the football match; the events include: pass, take a ball, shoot, goal, corner ball, free kick, point kick, out of bounds, goal ball, salvage, free from bounds, put out of bounds, go out of sideline, go out of bottom line, foul, red card, yellow card;
the commentary generation system generates commentary texts according to the game events and the related information; the explanation text comprises player team introduction, player team news abstract text, match event explanation text and end words;
and the speech synthesis system generates the comment audio by using a speech synthesis model to the output comment text.
10. An automatic commentary device for football match according to claim 9, characterized in that the device performs a method according to any one of claims 2 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110597834.5A CN113268515A (en) | 2021-05-31 | 2021-05-31 | Automatic explanation device and method for football match |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110597834.5A CN113268515A (en) | 2021-05-31 | 2021-05-31 | Automatic explanation device and method for football match |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN113268515A true CN113268515A (en) | 2021-08-17 |
Family
ID=77233597
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110597834.5A Pending CN113268515A (en) | 2021-05-31 | 2021-05-31 | Automatic explanation device and method for football match |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113268515A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114491143A (en) * | 2022-02-12 | 2022-05-13 | 北京蜂巢世纪科技有限公司 | Audio comment searching method, device, equipment and medium for field activity |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105304080A (en) * | 2015-09-22 | 2016-02-03 | 科大讯飞股份有限公司 | Speech synthesis device and speech synthesis method |
| CN105468607A (en) * | 2014-08-26 | 2016-04-06 | 腾讯科技（深圳）有限公司 | Generation method and device of live picture |
| CN108228869A (en) * | 2018-01-15 | 2018-06-29 | 北京奇艺世纪科技有限公司 | The method for building up and device of a kind of textual classification model |
| CN109635103A (en) * | 2018-12-17 | 2019-04-16 | 北京百度网讯科技有限公司 | Abstraction generating method and device |
| CN110769178A (en) * | 2019-12-25 | 2020-02-07 | 北京影谱科技股份有限公司 | Method, device and equipment for automatically generating goal shooting highlights of football match and computer readable storage medium |
| CN110971964A (en) * | 2019-12-12 | 2020-04-07 | 腾讯科技（深圳）有限公司 | Intelligent comment generation and playing method, device, equipment and storage medium |
| US20200294365A1 (en) * | 2019-03-12 | 2020-09-17 | Fayble, LLC | Systems and methods for generation of virtual sporting events |
| CN112423093A (en) * | 2020-10-16 | 2021-02-26 | 腾讯科技（深圳）有限公司 | Game video generation method, device, server and storage medium |
| CN112766546A (en) * | 2021-01-05 | 2021-05-07 | 上海大学 | Machine learning-based football match win and loss real-time prediction method |
-
2021
- 2021-05-31 CN CN202110597834.5A patent/CN113268515A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105468607A (en) * | 2014-08-26 | 2016-04-06 | 腾讯科技（深圳）有限公司 | Generation method and device of live picture |
| CN105304080A (en) * | 2015-09-22 | 2016-02-03 | 科大讯飞股份有限公司 | Speech synthesis device and speech synthesis method |
| CN108228869A (en) * | 2018-01-15 | 2018-06-29 | 北京奇艺世纪科技有限公司 | The method for building up and device of a kind of textual classification model |
| CN109635103A (en) * | 2018-12-17 | 2019-04-16 | 北京百度网讯科技有限公司 | Abstraction generating method and device |
| US20200294365A1 (en) * | 2019-03-12 | 2020-09-17 | Fayble, LLC | Systems and methods for generation of virtual sporting events |
| CN110971964A (en) * | 2019-12-12 | 2020-04-07 | 腾讯科技（深圳）有限公司 | Intelligent comment generation and playing method, device, equipment and storage medium |
| CN110769178A (en) * | 2019-12-25 | 2020-02-07 | 北京影谱科技股份有限公司 | Method, device and equipment for automatically generating goal shooting highlights of football match and computer readable storage medium |
| CN112423093A (en) * | 2020-10-16 | 2021-02-26 | 腾讯科技（深圳）有限公司 | Game video generation method, device, server and storage medium |
| CN112766546A (en) * | 2021-01-05 | 2021-05-07 | 上海大学 | Machine learning-based football match win and loss real-time prediction method |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114491143A (en) * | 2022-02-12 | 2022-05-13 | 北京蜂巢世纪科技有限公司 | Audio comment searching method, device, equipment and medium for field activity |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Shih | A survey of content-aware video analysis for sports | |
| Tejero-de-Pablos et al. | Summarization of user-generated sports video by using deep action recognition features | |
| Merler et al. | Automatic curation of sports highlights using multimodal excitement features | |
| JP2008511186A (en) | Method for identifying highlight segments in a video containing a frame sequence | |
| Hosseini et al. | Fuzzy rule-based reasoning approach for event detection and annotation of broadcast soccer video | |
| US20100005485A1 (en) | Annotation of video footage and personalised video generation | |
| US20210394060A1 (en) | Method and system for automatically generating video highlights for a video game player using artificial intelligence (ai) | |
| Hari et al. | Event detection in cricket videos using intensity projection profile of Umpire gestures | |
| Chiu et al. | Tagging webcast text in baseball videos by video segmentation and text alignment | |
| Khan et al. | SPNet: A deep network for broadcast sports video highlight generation | |
| Kang et al. | An E-sports video highlight generator using win-loss probability model | |
| CN113268515A (en) | Automatic explanation device and method for football match | |
| US8300894B2 (en) | Method for decomposition and rendering of video content and user interface for operating the method thereof | |
| KR102702069B1 (en) | Method of controlling sports activity classification learning apparatus, computer readable medium and apparatus for performing the method | |
| WO2021240679A1 (en) | Video processing device, video processing method, and recording medium | |
| US11790654B2 (en) | Derived competition | |
| Lie et al. | Combining caption and visual features for semantic event classification of baseball video | |
| Wang | A Deep Learning Algorithm for Special Action Recognition of Football | |
| Choroś | Video structure analysis and content-based indexing in the Automatic Video Indexer AVI | |
| CN110314368B (en) | Auxiliary method, device, equipment and readable medium for billiard ball hitting | |
| US20240001242A1 (en) | System for recognizing player behavior and game situation in sports game video | |
| Maram et al. | Images to signals, signals to highlights | |
| Choroś et al. | Automatic detection of play and break segments in basketball videos based on the analysis of the slope of the basketball court boundary | |
| Yamashita et al. | LSTM-Based Prediction of Ball Trajectory for Automatic Foul Judgements in Kin-Ball Sport | |
| JP7552155B2 (en) | Image processing device, image processing method, and model generation device |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20210817 |