Zhejiang University Scientists "Dream Linkage" China Table Tennis Team in the Olympic Games
Will table tennis bouncing on the blue table, football rolling on the pitch and badminton flying one day become "close partners" with artificial intelligence?
At the recently concluded Tokyo Olympic Games, the China table tennis team scored four gold medals and three silver medals, which also made a scientific research team behind it, the Zhejiang University Sports Big Data Innovation Team, appear.
What different landscapes have this innovative team brought to competitive sports by using artificial intelligence? Recently, the reporter went to the State Key Laboratory of CAD&CG of Zhejiang University to find out.
Unexpected intelligent ping-pong
On July 29th, in Tokyo Gymnasium, the semi-final of women’s singles table tennis was in full swing. You wouldn’t expect that besides the athletes, referees and few spectators in the stands, there is also a powerful artificial intelligence platform monitoring the game.
Every serve, swing, move, etc. on the field is recorded by this AI cloud platform deployed in Tokyo. These data are transmitted to the technical team of "table tennis intelligent big data analysis platform" 2442 kilometers away from the gymnasium at an average speed of 100 megabytes per second.
Almost at the end of the game, a technical and tactical analysis report about the game was immediately pushed to the tablet computers of the coaches and athletes of China table tennis women’s team by the AI cloud platform. These latest game videos and technical and tactical data analysis reports are directly used for the follow-up preparations of China table tennis women’s team.
The reporter learned that technical and tactical data used to be collected by watching videos after the game. However, because the speed of table tennis is too fast, researchers often can’t quickly, accurately and comprehensively mark all the athletes’ game behaviors, such as players’ hitting skills, landing points, positions, racket sequences and other information. It sometimes takes about 5 hours to collect detailed data of a game and make in-depth and detailed analysis.
This work is not only time-consuming and laborious, but also very challenging.
How to delete redundant relay shots, accurately identify the score changes of each round, detect and locate each beat, and efficiently label the data automatically or semi-automatically, has become a scientific research difficulty.
Deng Dazhen is a fourth-year doctoral student in the State Key Laboratory of CAD&CG, School of Computer Science, Zhejiang University, and a member of the sports big data innovation team of Zhejiang University. Deng Dazhao told reporters that in order to solve this problem, he and two classmates spent three months building a special model and realizing round segmentation, and developed an interactive data labeling method in the following year, successfully achieving high-quality data collection.
Professor Wu Yingcai, the doctoral supervisor of Deng Dazhao and vice president of the School of Computer Science and Technology of Zhejiang University, is the main developer of this intelligent big data analysis platform for table tennis that can "update combat effectiveness in real time". On the front line of AI cloud platform, it is Professor Zhang Hui, director of the Department of Physical Education, School of Education, Zhejiang University, who makes further analysis and feeds back key information to coaches and athletes in time. The two of them are also key figures in the sports big data innovation team of Zhejiang University. With the support of the State Sports General Administration and the Chinese Table Tennis Association, this team has been providing services for the national table tennis team in terms of game data analysis and technical and tactical research.
Since 2018, Wu Yingcai led the team to start the annotation analysis and platform construction of table tennis big data. Now, the sports big data innovation team has developed an intelligent big data analysis platform for table tennis based on interactive visual human-computer interface, combined with artificial intelligence algorithm and human experience wisdom, and realized semi-automatic data annotation, analysis and presentation.
This platform continues to make technological breakthroughs, which will shorten the labeling time. 2 hours, 1 hour … Now, the data can be collected as soon as the game is over.
Dr. Zhou Zheng from the Department of Physical Education, College of Education, Zhejiang University, and graduate students and undergraduates majoring in table tennis are responsible for the collection and analysis of competition data. Zhou Zheng and his mentor Zhang Hui have been deeply involved in the training and competition of the national table tennis women’s team for a long time. In addition to the Tokyo Olympic Games, they also participated in the series of competitions of the national table tennis team in preparation for the 2019 World Championship and World Cup, and produced a large number of targeted technical and tactical analysis cases and statistical reports, which contributed to the excellent achievements of the China table tennis team.
Now, the Tokyo Olympic Games has ended, and the China table tennis team has once again proved its strength, which also makes the scientific and technological forces behind it shine brilliantly on the world stage.
From evaluating technology to formulating tactics
Surprisingly, the table tennis intelligent big data analysis platform can not only evaluate technology, but also formulate tactics. This opens up a new direction for human-computer interaction.
The reporter noted that the application of artificial intelligence to the field of table tennis is nothing new. In 2011, two humanoid robots "Wu" and "Kong" developed by the Robotics Laboratory of the Institute of Intelligent Systems and Control of Zhejiang University were officially unveiled. These two robots that can play table tennis can capture the movement track of table tennis in the air through the camera mounted on their heads, predict the impact point of the ball, and then make corresponding response actions, and even achieve a fight with people. The table tennis intelligent big data analysis platform developed now is mainly aimed at the training and competition of excellent athletes, helping coaches and athletes to better observe, understand and analyze the technical and tactical advantages and disadvantages of their opponents and themselves, so as to know ourselves and ourselves.
It turns out that an important aspect of table tennis technical and tactical analysis is to judge the quality of players’ hitting. This requires coaches and researchers to manually evaluate the quality of athletes’ shots on the spot or when watching the game video. Another question arises from this: what to do when the number of competitions is very large; If there are 100 or even 1000 videos, can such a task be completed?
Therefore, evaluating the quality of swing has also become a goal that sports big data innovation teams need to achieve.
The first thing they need to solve is how to let the computer learn the knowledge of professional table tennis players, and combine a series of attributes such as the technology used in each beat, the hitting position and the landing point to objectively and accurately evaluate the player’s hitting process. In order to solve this technical problem, from last year, Wu Yingcai began to tackle key problems through team cooperation with Professor Zhou Zhihua of Nanjing University. This year, they finally made important progress and original achievements.
"We use the theoretical framework of deductive learning to integrate more than 30 table tennis technical and tactical rules into data-driven machine learning, and develop a framework that can automatically evaluate the quality of swing." Wu Yingcai said that deductive learning can perfectly integrate data-driven machine learning technology and rule-driven artificial intelligence technology. Now, as long as the game video is input on the analysis platform, the hitting quality of each shot of the athletes in the video can be automatically obtained.
By overcoming the difficulties step by step, there are more than 8,000 high-precision competition data of international competitions on the table tennis intelligent big data analysis platform. Since 2018, the research team has started the simulation prediction of the game based on these data, and truly upgraded from technology to tactics. "With big data and artificial intelligence, we can simulate and deduce to predict how the winning percentage will change when our players change a certain skill and tactics." Wu Yingcai said.
This idea has also been applied to badminton. Xie Xiao, a special researcher in the Department of Physical Education, Zhejiang University, demonstrated a new system being developed by the team.
As long as you wear a VR helmet, you can see the tracks left by athletes waving badminton rackets on the court. At the end of a game, all the action tracks are presented on the computer screen, as if countless "ribbons" were thrown in the air. These "ribbons" are grouped and classified and distinguished by color.
This is Shuttlespace, an immersive analysis system that can help experts analyze badminton data. Xie Xiao said that in the traditional analysis, the video of badminton match is separated from the analysis chart, which makes people unable to intuitively perceive the real situation on the court. Shuttle Space uses virtual reality technology to simulate the three-dimensional trajectory of man and ball in the most realistic form, which is enough for experts to perceive and analyze the data from the player’s subjective perspective and better formulate tactics.















