To assist gymnastics judges in their decision-making process, Fujitsu developed a GPU-accelerated deep learning application that can sense the movements of athletes and analyze them as scoring data.
“[This is] a big step towards the future,” said, Morinari Watanabe, president of the International Gymnastics Federation “Scoring controversies must become a thing of the past, and the technology that Fujitsu has been developing will reinforce trust in judgment.”
The system was first used at the Artistic Gymnastics World Championships in Stuttgart, Germany earlier this month.
With the help of 3D sensors and computer vision, the system allows judges to view gymnast’s performance in 3D from various angles, helping support fair and accurate scoring.
At the world championships, Fujitsu installed 30 AI-equipped systems that tracked the movements of the 547 gymnasts from 92 different countries. Prior to the competition, the company digitally scanned each gymnast competing in Stuttgart, to enhance the accuracy of the system.
During the competition, data was fed from the sensors to the deep learning workstation, equipped with an NVIDIA TITAN GPU, with cuDNN-based software for inference, that measured and analyzed speeds, angles, and skeletal positions of the athletes in real time.
The International Gymnastics Federation sees significant potential in the technology, not only in the matter of judging but also in the future for education, coaching, and training,” the organization said.
“The data provided might help to enhance gymnasts’ movements or correct body posture, and thus aid in lowering the risk of injuries,” the organization wrote in a blog post.
Fujitsu says they plan to evolve the application beyond judging support, including for athlete training and entertainment applications.