Video mining and sports analytics
I was one of those people who were blown away by Deb Roy’s TED presentation, where he showed how he had collected 90,000 hours worth of home video footage and mined that footage to find out exactly how his son developed language skills. Some commenters remarked that the actual amount of knowledge gained through this exercise was not that impressive. That may be so, but I still thought it was remarkable how Roy’s MIT Media Lab team had managed to turn all of that video data into gorgeous visualizations that made intuitive sense (like the “3D density plots” of word usage throughout the house, for instance). There has to be a serious software infrastructure somewhere in there to enable this kind of analysis. Roy has a company, Bluefin Labs, and I came across an intriguing press release from 2009 which stated that Bluefin would start to “index video” for the consumer sports market, so that sports videos could be easily searched and analyzed. However, it seems Bluefin has since dropped the idea, as their home page now talks about “measuring consumer response” through digital channels rather than sports video analytics.
However, there are other companies that have taken up the idea. A recent Wired article (Hoops 2.0: Inside the NBA’s Data-Driven Revolution) describes a system called SportsVU, which uses video cameras to track the players and the ball to a remarkable level of detail. The technology grew out of missile-tracking applications and optical recognition algorithms to that. It was first applied to soccer, but it was later decided that basketball would be more lucrative. With the SportsVU system – and, not to forget, the analytics to crunch the raw image data – it’s possible to track some pretty complicated metrics, as illustrated in this example for Kevin Martin.
A different way to quantify basketball player performance is used by Infomotion Sports, as descibed in this Yahoo! Sports article. Infomotion’s 94Fifty technology, instead of using video cameras, actually measures the movements of the ball directly using sensors embedded into the inside of the ball. Through “feeling” the ball’s motion, one can quantify a lot of things about a player, such as dribbling speed, shot angles and spin. A quote from the article:
“A coach can tell me that [a player] needs to work on his left hand,” Kamil said. “[But] we can tell you that his right hand is 14 percent more dominant than his left.”
Of course, a lot of people would be skeptical as to how much you really gain from this kind of data-driven approach compared to the good old gut feeling of a coach. I count myself among the skeptics, but I think the technology is intriguing.