This is a review of the computer vision and machine learning research conducted by Omar Ajmeri and Ali Shah.
NFL coaches are always looking for weaknesses in their opponents. They spend countless hours going through game film, searching for ways to increase their chances of winning. The process is tedious and often results in many errors. Scouting your own team is very straight forward but scouting all of your opponents is a very time consuming task. Computer vision and machine learning techniques can help.
The classification of NFL “All-22” film has been automated from start (offensive formation labeling) to finish (video player tracking coordinates throughout the life of a play).
A wealth of data can be generated from the resulting system that can help us see coaching tendencies. Analysis of the most common formation, Singleback Ace Pair Slot (two tight ends on the right side of the line, two wide receivers to the left of the line, with one running back) generated the following information: (1) equally likely to pass or run out of this formation; (2) 65% of runs were to the right side, equally split between right guard and right end, with few runs up the middle; (3) 81% of passes were on short routes; and (4) passes gained an average of 4.6 yards per play, while runs averaged just 2.5 yards per play. This is just the beginning of what is possible, giving defensive coordinators and players the ability to create a stronger game plan against their opponents.
Another factor that can be analyzed is player tracking. Looking at average speed and acceleration throughout a game can help coaches gain a better understanding of player fatigue which would allow them to develop game plans that would take this into account.
It is also possible to gain a clearer understanding of how effective players are at running different patterns. Coaches will be able to compare how fast different players are able to run the same type of pattern, allowing for a side-by-side comparison. This would be useful when looking at possible trades, free-agent signings and contract negotiations.
This classification process has the potential to save coaches hours of time that could be spent on more productive activities. At the same time it will give coaches a greater insight into their opponents, allowing them to design a more effective game plan. It will positively affect game planning, scouting, and provide better evaluation of individual players and coaches. The ability to analyze masses of player location data in a short period of time will change how football coaches scout and analyze players and opposing coaches through the league.
This classification process could be extended to replicate the analysis for defensive formations and player analysis. With an analysis of both offensive and defensive formations coaches could examine the relationship between offensive and defensive play calling. The process could also be expanded to include kickoff and punt coverage formations.
How mature is your team’s analytics program? Take the Sports Analytics Maturity Assessment.
Learn about the Groundbreaking Sports Analytics Model coaches and sports analysts are talking about!
Learn all about Sports Analytics here.