Home Sports Preparing for the Winter Olympics, artificial intelligence technology shows off

Preparing for the Winter Olympics, artificial intelligence technology shows off

by admin

Original title: Preparing for the Winter Olympics, artificial intelligence technology shows off

In the women’s throwing competition of the Tokyo Olympic Games held in early August, Chinese players Gong Lijiao and Liu Shiying performed well and won the shot put and javelin gold medals. Gong Lijiao set a personal best with 20 meters 58 and also won the first gold medal for the Chinese team in Olympic field events.

After the game, the Chinese Athletics Association sent a letter of thanks, thanking Beijing Sport University for using science and technology to find a “key way” for Chinese throwers in the technical link to achieve self-breakthrough.

“We use artificial intelligence technology to analyze the athletes’ movement techniques, put forward suggestions for improvement, and use scientific and technological means to help the athletes.” said Professor Liu Hui, Dean of the Institute of Sports and Health, Beijing Sport University. The reporter learned from interviews that this technological system that emerged in the Tokyo Olympics is currently serving the continuous improvement and upgrading of Chinese athletes who are actively preparing for the Beijing Winter Olympics.

Break through the limitations of traditional motion capture methods

The use of biomechanical methods to study human movement requires quantitative analysis of the actions performed, and the basic premise is inseparable from data.

Fast and high-quality acquisition of athletes’ movement technical data is a key bottleneck that needs to be overcome urgently. Traditional motion capture technology requires either fixing reflective markers or inertial sensors on the human body, or manually identifying human joint points.

See also  Italy-Slovenia 4-0, azzurrini in the European U.21 finals

“The former cannot be used in competitions, and the latter is due to heavy workload, time-consuming, and poor repetitiveness, which seriously affects the feedback speed and reliability of action technical analysis and limits the application of biomechanics in assisting competitive sports.” Liu Hui explained. road.

How to break the predicament? Liu Hui’s team used artificial intelligence technology based on deep learning principles to establish a neural network model to realize computer automatic recognition of human body joints in action videos, and then establish a computer system suitable for competitive sports and general biomechanics research-non-reflective An artificial intelligence system that automatically captures human movements.

As the person in charge of the scientific and technological winter Olympics key special project “Winter sports athletes’ special ability characteristics and scientific selection of key technology research”, Liu Hui told the reporter of Science and Technology Daily that the system has been applied in the training of national speed skating and cross-country skiing events, and has won more than 8,000 The technical data of the person-time movement of the game makes machine deep learning more “handy.” For skaters and skiers’ motion capture and technical analysis, it can not only be accurate to specific details, but also quickly feedback the analysis results.

Multiple algorithm technologies ensure fast and accurate automatic identification

In the research, Liu Hui’s team found that automatic analysis of motion video needs to solve at least three problems: “can follow”, “accurate recognition” and “high precision”.

The scene of sports shooting video, the picture environment is complex and diverse. The research group combines optical flow tracking technology in the commonly used moving human tracking algorithm, that is, the main ID (identity person) is accurately locked through the amount of movement and the magnitude of the movement, which can effectively avoid image blur caused by fast movement and reduce complexity. The background and other factors interfere to ensure that they can “follow up”.

See also  Watford, bench for Claudio Ranieri - Sport - Football

At the same time, machine learning is performed on a large number of marked training data, and a neural network is formed by a computer system, which can identify the joint points of the human body in different motion postures, and achieve “recognition accuracy”.

In addition, the system has the function of independently calculating the joint points of each frame of image. How to reduce the random error of the joint point position during independent calculation? Liu Hui introduced that the algorithm is used to increase the time constraint for continuous motion, that is, the high-frequency error of each joint point is identified and eliminated, so as to correct the position and coordinates of the joint point, and finally obtain a high-precision calculation result.

“Since 2019, after several versions of iterative upgrades, the system has been able to quickly and accurately identify the human body joint points in sports videos, and can also perform better automatic recognition of human movements such as rotation and roll.” Liu Hui In other words, the system synthesizes and outputs the three-dimensional coordinates of all recognition points, and supports batch automatic analysis and index calculation of multiple videos.

Liu Hui told the reporter of Science and Technology Daily that if the system uses industrial video recorders, data transmission and processing can often be completed in 1-3 minutes. “This will play a vital role for the skill athletes to experience the competitive state and master the technical essentials.” She said.

High-altitude motion capture is no longer difficult

It is understood that this system also provides a variety of spatial three-dimensional calibration solutions, which can solve the data collection problem of large-scale and high-altitude movements.

See also  "I Really Love You" Xiao Yan kindly helped Hao Dawei design Cheng Haonan_YNET.com北青网

What is the scope of this? Liu Hui said that the space range of 20-30 meters can be covered. Especially for ski jumping aerials, it can find room for improvement for technical details that are difficult for coaches to recognize with the naked eye.

At present, the system has been used in the preparation and training of the national team for steel-framed snowmobiles, figure skating, and ski jumping. It will provide important scientific and technological support for athletes to prepare for the Beijing Winter Olympics.

Editor in charge: Meng Xiangyu (EN009)


0 comment

You may also like

Leave a Comment

This site uses Akismet to reduce spam. Learn how your comment data is processed.

This website uses cookies to improve your experience. We'll assume you're ok with this, but you can opt-out if you wish. Accept Read More

Privacy & Cookies Policy