The NFL and Amazon Web Services, Inc. (AWS) announced the results of its artificial intelligence competition, which challenged data scientists to teach computers to automatically detect players involved in head impacts from NFL game footage. The new computer vision models further strengthen the data and insights at the heart of the NFL's effort to understand and reduce head injuries.
The NFL reviews game footage of all major injuries, analyzing each injury from every angle, recording 150 different variables. The winners' models automate that process, making review more comprehensive, accurate and 83 times faster than a person conducting the analysis manually.
Insights from the data will be used to enhance the NFL's injury-reduction efforts, which include driving innovation in protective equipment design and improvements to coaching and training strategies and player protection-based rules changes.
"The innovative ideas brought to this competition from data scientists around the world will be transformative, driving a staggering improvement in accuracy of computer vision models over just a three-month competition," said Jennifer Langton, SVP, NFL Player Health and Innovation. "The success of this challenge speaks to the power of the crowdsourcing model that the NFL has deployed over the last decade to drive innovation in player health and safety, and we are thrilled to have had some of the brightest minds in data science from around the world working on our challenge."
More than 1,000 data analysts from 65 countries competed in the challenge and five winning models were awarded a total of $100,000. Kippei Matsuda from Osaka, Japan took the top prize of $50,000, followed by Takuya Ito from Tokyo, Japan who received $25,000. Third-, fourth- and fifth-place finishers received $13,000, $7,000, and $5,000, respectively. The NFL gave data scientists access to NFL game data and challenged them to build on last season's competition, which had crowdsourced models to detect helmet impacts from NFL game footage, to now be able to automatically identify the specific players involved in each helmet impact.
"This was the most exciting competition I've ever experienced," said Kippei Matsuda, who finished in first place. "It's a very common task for computer vision to detect objects in 2D images, but this challenge required us to consider higher dimensional data such as the 3D location of players on the field. I would be honored if my AI can help improve the safety of NFL players."
These new computer vision models will also help the NFL and AWS continue to build the "Digital Athlete," a virtual representation of an NFL player that can be used to better predict and help prevent injuries to players. The Digital Athlete will ultimately help the NFL and its clubs develop individualized training and recovery regimens, conduct real-time risk analysis for injury during games, and identify and evaluate additional player protection initiatives.
"AWS and the NFL are fostering an understanding of how to treat and rehabilitate injuries in the near term and eventually predict and prevent injuries in the future by leveraging data," said Dr. Priya Ponnapalli, senior manager at the Amazon Machine Learning (ML) Solutions Lab. "New computer vision models developed in this challenge, and the hard work put in by all the teams involved, bring us steps closer to our goal and I couldn't be more thrilled to see how this work transforms the sport in the coming years."