Computer vision is having a sort of coming-out moment. (And more broadly, the umbrella of machine learning that it falls under.) Everyone in the field got really excited when a computer beat the world champion at a game called Go. Every tech company seems to be devoting more resources to acquiring promising start ups. And self-driving cars seem poised to introduce a pretty radical change to the car-owning culture of the U.S.
Getting under the hood a bit, a lot of the applications of machine learning have to do with classification and response. A chatbot reads your message, prepares a suitable response, and serves it back. A car sensor system diagnoses the current situation and then adjusts the car’s motion accordingly. But these headlines represent a bit of an evolutionary step past what we think of as video analysis in sports. Video analysis in sports can generally rest easy with the classification step only.
The most successful sport in utilizing computer vision and video analysis has been the NBA, by far. The Sport VU system is installed in every NBA arena, where it can capture video of every second of every NBA game. This video is made up of millions of snapshots that include all 10 players and the ball. By tracking the position of these 11 objects, Sport VU enables all kinds of measures: the distance from a defender to a shooter, the location from which every shot was taken, and by who. This is an incredibly short list of what it can do. And because teams have access to the data, the analyses they can do with their ever-expanding teams of data scientists is limited only by their imagination and the estimated competitive advantages.
Other sports, not so much. Baseball has just begun to use this sort of analysis to measure runner speed, outfielders’ path to fly balls, launch angles, etc. Football just released their full-field camera angles a few years ago, but as of yet, any substantial progress in video analytics is either kept under wraps or nonexistent. Hockey seems to be a similar story. I think it’s telling that golf is probably the sport beyond basketball that has progressed farthest in the use of video; and for them, it’s more about the viewer experience.
So why basketball (and by extension, none of the other sports)? A few hypotheses:
- There is less player to player contact than some sports (compared to football or rugby where it’s more of a scrum), meaning that players are generally moving freely. This means that from a video perspective, it’s possible to distinguish individuals with more accuracy.
- The ball is large and easy to see with a camera. Hockey felt the need to have a blue tracer line follow the puck around for a year or so; it’s not much easier for a camera to track it.
- There are relatively few players on the court at one time especially compared to football; this means that like point #1, it’s easier for a camera to track individuals around the court.
- The team works as a whole in a more flowing game style, compared to the individualized discrete events of baseball; this means that it’s very difficult to accurately paint a picture of everyone’s contribution, producing a necessity for this analysis that baseball does not have.
What does that all mean for lacrosse?
On the positive side, the flowing style makes it very difficult to quantify a player’s impact using the available counting stats. We can observe that someone is a good defender, but comparisons are difficult to quantify, and any time a subjective visual observation can be replaced with an objective measurement, that’s a good thing in our world (Grantland Rice just turned over in his grave). Also, there is a similar distribution of players around the field as in basketball, at least when you adjust for the size of the field. While it’s not a non-contact sport, the players don’t have too much equipment on, which means that from a computer vision perspective, it’s not all that difficult to isolate all 14 players.
On the negative side of the ledger, the ball is tiny, white, and obscured by a stick head 95% of the time.
The other thing holding lacrosse back: money. Or to be more accurate, ROI. To date, there has not been the sort of analytics community in lacrosse that there has been in other sports. This means that we are a very long way from critical mass, which means that the techniques/skills needed to do this work have not been focused on lacrosse. (It’s fascinating how in other sports, the amateur data scientists published online, and eventually got snapped up by the front offices themselves.) I think that a lack of a popular professional layer to lacrosse hurts too.
Take heart though, the supply of data scientists is growing, the cost of the technology and techniques to support analysis like this is falling, and the popularity of the sport is growing. We’ll get there.
So what would that look like? What would it mean for lacrosse to jump on video analysis?
First of all, we would know a lot more about the mechanics of defensive rotations. This is such a critical part of defense, but to date, it’s about a coach’s philosophy, which may or may not be adjusted to account for personnel. But like in most areas of lacrosse, the best we can do at quantifying a team’s effectiveness is to ask the experts who have watched the teams. Better data on this has the potential to fundamentally alter how teams play defense, or at least how they adjust their defensive strategies.
Second, we would have a much better grasp for the true impact of scorers and non-scorers. Currently, we have a very small number of very broad counting stats. On offense, goals and assists tell us what happened at the very end of a scoring sequence, but they really can’t tell us much about the impact a dominant offensive player has in terms of extra attention and how that can open up opportunities for teammates. On defense, it’s even more difficult. Video analytics could create the potential for a WAR / PER metric for lacrosse.
Lastly, (and forgive me for broadness here) we would have the opportunity to create a better product. I am generously putting myself into the “lacrosse community” here, but as lacrosse continues to spread, the quality of the play needs to increase with it. More video means more understanding of the game, and that can lead to non-blue bloods having a better shot at success. College basketball tell us that if nothing else, people love a good underdog story. Perhaps video analysis can be a field-leveler in a way that makes them more common.