Nylon Calculus 101: Intro to SportVU

By now, if you follow developments in NBA technology, statistics or even broadcasting, you’ve probably heard of either SportVU, “player tracking stats,” or both. But just what is SportVU and what does it do?

The System

SportVU is a proprietary camera system owned by STATS, LLC.[1. The official stats partner of the NBA and friend of the TNC program. STATS is in charge of all the raw goodies that makes stats.nba.com such an ever-expanding treasure trove of data and information.] Based on technology developed for ballistic missile tracking, SportVU was first introduced as a way of capturing the movements of players on European soccer fields. STATS acquired the technology, re-designed the system to better fit basketball play and NBA arenas and began marketing the system to teams. After being adopted by individual organizations over several years, the NBA partnered with STATS prior to the 2013/14 season to install the system in all 29 arenas. This coming[2. 2015/16 at the time of this writing.] season will thus be the third with complete coverage of the league by SportVU.

The system itself consists of 6 small, high-tech cameras mounted in the rafters of each arena. These cameras are tied to a computer which in turn plots the position of all ten players and the ball 25 times per second. The system is managed in the arena by two operators on hand both to ensure the system is working, but also to supply additional information when necessary. The system id’s players by jersey numbers, so occasionally will require input to assist that identification should a those numbers be obscured from the cameras by a mass of bodies congregating in one area of the court.

The information recorded includes about 800,000 data points from each game, and roughly 1 billion over the course of a full regular season. That’s actually the easy part.

The Stats

The raw data produced by SportVU is essentially nothing more than a series of moving dots. Turning those dots into useful basketball statistics is where the actual rocket scientists get involved[1. Remember, ballistic missile tracking…]. Virtually any[2. Not really any due to technical limitations of the optical tracking technology, more on that in a second.] event which takes place on a basketball court can be tracked, categorized and counted for every game in the SportVU system, so long as you know what you are looking for in that mass of moving dots. A certain pattern of movement is a dribble. Another is a pass. Yet a third is a shot.

From their, more complex interactions can be categorized such as drives to the basket, assist chances, and contesting shots at the rim. And voila, new stats!

So What?

Many of the initial information released to the general public from the SportVU system seem rather pedestrian. Haven’t coaches been tracking this stuff for years? While yes, much of this information is exactly the kind of thing a coach might like to know, and assign an assistant to chart it. However, with human charting comes the possibility of errors, inconsistency but most importantly time constraints. No coach in the world has the patience to meticulously track every pull-up jump shot from every player in the league over a season[2. There were just under 50,000 pull-up jumpers taken in 2014/15.]. If the rocket scientists do their job, nobody has to try. And even if the initial coding is wrong, it’s fixable with a change in formula, rather than having to go through the same film for a second time to correct earlier errors. As Rajiv Maheswaran, head of leading NBA analytics consulting firm Second Spectrum told me, the system “might not see a game as well as a coach, but it can see all the games at once.”

This ability to see finer details from every single game[4. With a few exceptions, the system was inoperable for various reasons for four games in 2014/15, and no tracking data was captured from those contests.] allows for looking at how closely every single shot is defended, who runs the most on average every game, and which players shoot best off the dribble as well as near limitless other topics.

So How Does This All Help?

It’s hard to say confidently, because the information presented to the public on NBA.com is the merest tip of the iceberg. The information teams can pull out of the full data set is limited only by their imaginations, and their available programming talent. Of course, both can be pretty sizable limitations.

To handle the programming side, STATS itself offers a package of predeveloped stats and reports to teams, while other groups such as the aforementioned Second Spectrum bring their own PhD-level data science to bear. A favorite example at present is Maheswaran’s claim that Second Spectrum can properly identify over 90% of the pick-and-rolls which occur in an NBA game, and can then sort, collate and aggregate them by the players involved, the outcomes of each play and numerous other factors which allow for more informed decision about such topics as defensive strategy and player personnel decisions.

So It’s Already Changed the NBA World?

Yes and no. Through SportVU, we know more than we knew two years ago. But much of that learning has been realizing other things we just didn’t know. The two seasons of coverage of the full league is really a blink of the eye, simply because so many questions haven’t been asked of, let alone answered from the SportVU data.

But What Don’t We Know

In addition to the questions left unasked, there are some limitations of the technology itself. The system measures players by the position of their torsos. As such, it can’t see hand position, so there is not quite yet a good test for “hand down, man down[5. I’m so, so sorry.].” Nor can the system tell where players are looking or pick up on vitally important information which a player on the floor might pick up, like which way is a defender leaning or has a player signaling for a lob while cutting to the rim? Finally, the system doesn’t and probably can never know play calls or suss out another team’s scheme or other similar “technical” basketball elements.