Machine Vision: The Future of Player Tracking Systems with Hudl

The new­ly-released Hudl Pro Suite has been devel­oped to cap­i­talise on mod­ern devel­op­ments in video and opti­cal track­ing tech­nolo­gies to pro­vide new data ser­vices to the pro­fes­sion­al sports market. 

Player track­ing sys­tems were first intro­duced in foot­ball dur­ing the late 1990’s. They were based on opti­cal track­ing, which required spe­cialised cam­era systems.

The lat­est tech­nol­o­gy allows these sys­tems to com­pete with satel­lite or local posi­tion­ing-enabled wear­ables (often referred to as GPS and LPS) as the tool of choice for gen­er­at­ing posi­tion­al and iner­tial data. Meaning they can quan­ti­fy phys­i­cal demands, auto­mate video telestra­tion, and pro­vide tac­ti­cal insights into team dynamics.

While wear­ables can pro­vide sup­ple­men­tary data, such as iner­tial and heart rate data, there are sev­er­al advan­tages to opti­cal track­ing tech­nol­o­gy. It’s non-inva­sive, more reli­able in sta­di­um envi­ron­ments and can pro­vide con­tex­tu­al data on both teams — a crit­i­cal aspect when judg­ing per­for­mance in a com­pet­i­tive environment.

Dr. Paul Neilson, Hudl’s for­mer direc­tor of mar­ket devel­op­ment, explained how the Hudl Pro Suite was developed. 

It all start­ed with us installing cam­eras for the tac­ti­cal feeds with­in sta­di­ums,” said Neilson. Then we thought, now we have these cam­eras in sta­di­ums, can we cre­ate a play­er track­ing sys­tem which can inte­grate with our exist­ing soft­ware solutions?”

We will pro­vide 25 coor­di­nates every sec­ond on a player’s posi­tion, that’s for the whole game.”

Hudl’s research and devel­op­ment team worked on this new project for the best part of two years to build a play­er track­ing sys­tem which utilis­es machine vision’ and machine learn­ing’ to track any object with­in an image frame.

Fundamentally, what our play­er track­ing sys­tem gives you is a wide range of data points on both play­ers and match offi­cials,” adds Neilson. We will pro­vide 25 coor­di­nates every sec­ond on a player’s posi­tion, that’s for the whole game.

So if you mul­ti­ply out 25 points of data per sec­ond in a game of foot­ball with 22 play­ers across 95 min­utes, you’re talk­ing over 3 mil­lion data points. That’s what we can cre­ate for a game of football.”

Once this data has been col­lect­ed, there are a num­ber of uses which can ben­e­fit the team.

For exam­ple, you can mea­sure phys­i­cal exer­tion in and out of pos­ses­sion, high inten­si­ty activ­i­ty dur­ing tran­si­tions and the vol­ume and type of accel­er­a­tions and decel­er­a­tions per­formed. In addi­tion, posi­tion­al data can now be used to iden­ti­fy and report deep­er tac­ti­cal insights on team dynam­ics, for exam­ple team shape, press­ing, defen­sive bal­ance, numer­i­cal dom­i­nance, play­er attrac­tion and cre­at­ing space.”

You’re talk­ing over 3 mil­lion data points. That’s what we can cre­ate for a game of football.”

To test the con­sis­ten­cy and accu­ra­cy of the data col­lec­tion meth­ods, Hudl con­duct­ed exten­sive and fre­quent inter­nal test­ing. These includ­ed super­im­pos­ing veloc­i­ty and accel­er­a­tion data onto match video to allow visu­al inspec­tion and ver­i­fi­ca­tion of the posi­tion­al data and cal­cu­lat­ed data outputs.

These visu­als pro­vid­ed expert oper­a­tors and engi­neer­ing teams an effec­tive method of inspect­ing posi­tion­al data for obvi­ous track­ing issues,” con­clud­ed Neilson.

Initial field based data val­i­da­tion test­ing took place at Danish Superliga club Brondby IF. To learn more about this test­ing and Hudl’s con­tin­u­al invest­ment in data valid­i­ty and reli­a­bil­i­ty, click here.