Baseball for the Thinking Fan

Login | Register | Feedback

btf_logo
You are here > Home > Baseball Newsstand > Baseball Primer Newsblog > Discussion
Baseball Primer Newsblog
— The Best News Links from the Baseball Newsstand

Thursday, May 16, 2019

Sarris: MLB moving from Trackman to Hawk-Eye tracking system – The Athletic

Maybe now we’ll be able to get some infield defensive metrics.

Five years after it was installed in an effort to measure the previously unmeasurable, radar-based player- and ball-tracking system Trackman looks like it’s on its way out as the technology of choice for Major League Baseball. Multiple sources inside front offices across the league confirm that Hawk-Eye’s optical technology — known to tennis fans as the basis of the automated serve tracking system — is currently being installed in baseball stadiums across the country for a two-month run-up that should end in a full change in technology for the 2020 season.

MLB offered no comment on the scope of the upcoming deal with Hawk-Eye. Hawk-Eye and Trackman representatives had not responded to requests for comment at the time of publication.

Multiple sources cited an improvement in accuracy as the main reason for the change.

Jim Furtado Posted: May 16, 2019 at 06:18 AM | 4 comment(s) Login to Bookmark
  Tags: pay site, technology

Reader Comments and Retorts

Go to end of page

Statements posted here are those of our readers and do not represent the BaseballThinkFactory. Names are provided by the poster and are not verified. We ask that posters follow our submission policy. Please report any inappropriate comments.

   1. calming him down with his 57i66135 Posted: May 17, 2019 at 12:14 AM (#5842979)
i don't know if anyone here develops machine learning algorithms, but this seems like the kind of situation where algorithmic estimation would be cheaper and easier to implement, with no noticeable loss in effectiveness. you might lose .2% in accuracy but that difference in precision doesn't seem very meaningful in this usage case.
   2. villageidiom Posted: May 17, 2019 at 10:14 AM (#5843023)
i don't know if anyone here develops machine learning algorithms, but this seems like the kind of situation where algorithmic estimation would be cheaper and easier to implement, with no noticeable loss in effectiveness.
The problem isn't cost or feasibility. It's all about effectiveness.

Trackman is radar augmented by a few cameras. The radar component sucks. The further you go from any radar system the effectiveness decays, maybe not exponentially, but effectively exponentially. Worse, it's a stationary radar, not one of these revolving things like you might think on a battleship or for weather, taking frequent readings in many directions. It's pointed up the middle, essentially at the pitcher. That means it's very effective for the old PITCHfx stuff, less effective for CF, and... useless? let's say mostly useless - down the lines. When two fielders cross each other, the system can't tell which is which. When the cameras are looking for a white object in motion it appears like a line segment, so when the ball crosses a foul line it sometimes thinks the foul line is the ball and loses track of the actual ball - which could be resolved if the radar system were actually pointed toward the lines and detected movement. But no.

Nearly all the Statcast fielding stats they've released depend on so little additional info - specifically, fielder positioning - from what we had before Trackman. The wealth of useful additional info it provides is coming from a region of 60 feet 6 inches. That's why we see so much more about spin rates, launch angle, and exit velocity than we see about the fielders.

The new system will be like a dozen cameras, which when combined with the computing power to triangulate it all should be able to track everything, and do it far more effectively than today. Whether Hawk-Eye is actually the right implementation of such a system, I've no idea.

I know above I'm making it sound like it's a tire fire today. I'm sure Tango and others are doing the best the can with the information they have, and the stuff we have available is damn good. My point is there's some fundamental stuff we don't have today, because Trackman can't do it.
   3. villageidiom Posted: May 17, 2019 at 10:33 AM (#5843039)
Let me use an example.

Per MLB,
Catch Probability represents the likelihood that a batted ball to the outfield will be caught, based on four important pieces of information tracked by Statcast. 1. How far did the fielder have to go? 2. How much time did he have to get there. 3. What direction did he need to go in? 4. Was proximity to the wall a factor?


1 and 3: If launch angle, exit velocity, and weather effects can be determined accurately, then basic physics can tell you where the ball will land. The only other piece of information you need is fielder positioning.

2: If launch angle and exit velocity can be determined accurately, then basic physics can tell you how long a ball will be in the air.

4: The only information needed beyond what's needed for 1 and 3 is the field dimensions, which is static and need not be measured by Trackman/Hawk-Eye.

So for catch probability, they can get some sense of all this without relying on any readings away from home plate other than weather and the fielder's starting position. You still need those, and need them to be reliable. But you don't need to track the fielder's movement, nor the ball. You don't need radar to know where the fielder started, and weather measurement is already a thing.

It's likely a better system if you can reliably track the ball, and the fielder(s). I think there's a lot more that can be done for fielding measurement (and improvement) than is possible with Trackman. Rich Hill owes his career to Brian Bannister looking at spin rates and release points and deducing that Hill had Kershaw-like stuff that could produce Kershaw-like results if used differently. Right now the level of detail on defensive tracking isn't enough to get there.
   4. calming him down with his 57i66135 Posted: May 17, 2019 at 12:46 PM (#5843115)
Trackman is radar augmented by a few cameras. The radar component sucks. The further you go from any radar system the effectiveness decays, maybe not exponentially, but effectively exponentially. Worse, it's a stationary radar, not one of these revolving things like you might think on a battleship or for weather, taking frequent readings in many directions. It's pointed up the middle, essentially at the pitcher. That means it's very effective for the old PITCHfx stuff, less effective for CF, and... useless? let's say mostly useless - down the lines. When two fielders cross each other, the system can't tell which is which. When the cameras are looking for a white object in motion it appears like a line segment, so when the ball crosses a foul line it sometimes thinks the foul line is the ball and loses track of the actual ball - which could be resolved if the radar system were actually pointed toward the lines and detected movement. But no.
most of that is an issue of computational complexity. because it's easier to validate the answer than it is to compute the answer, you can train an algorithm to cut out the computation.
It's likely a better system if you can reliably track the ball, and the fielder(s). I think there's a lot more that can be done for fielding measurement (and improvement) than is possible with Trackman.
likewise, for this, it seems like you could map the player locations onto the physics engine from some video game, then train a machine learning algorithm from a dataset you derive from that simulation.

people forget how much information we can glean from a still picture. dimensionality already exists in the current environment.. it's just hasn't recognized because people who build these models haven't spent enough time and/or effort to develop a wider knowledge base.

as an example, if you take the descriptive information that we already know (dimensions of the field, dimensions of the players, relative locations of various objects within the field), then layer a few 500+ year old design principles on top of 24 FPS 4K video -- use the vanishing point to simulate relative distances; use lighting and shadowing to add dimensionality -- that will give you a dataset that you can map onto a physics engine to find the answers that you want.


the same process can be used to turn 4K video into a fully interactive VR environment.

You must be Registered and Logged In to post comments.

 

 

<< Back to main

News

All News | Prime News

Old-School Newsstand


BBTF Partner

Dynasty League Baseball

Support BBTF

donate

Thanks to
Harry Balsagne
for his generous support.

Bookmarks

You must be logged in to view your Bookmarks.

Hot Topics

NewsblogThe Hall of Fame Case for Thurman Munson
(10 - 8:38pm, Dec 08)
Last: PreservedFish

NewsblogMiller, Simmons elected to HOF on Modern Era ballot
(1 - 8:35pm, Dec 08)
Last: asinwreck

NewsblogOT - 2019 NFL thread
(159 - 8:29pm, Dec 08)
Last: It's regretful that PASTE was able to get out

NewsblogChicago Cubs facing federal review over Wrigley Field access
(2 - 8:27pm, Dec 08)
Last: Walt Davis

NewsblogYankees Have Reportedly Made $245MM Offer To Gerrit Cole
(13 - 8:27pm, Dec 08)
Last: rconn23

NewsblogRyan Thibs’ Hall of Fame Tracker
(374 - 8:26pm, Dec 08)
Last: Meatwad

NewsblogThe Hall of Fame Case for Ted Simmons
(6 - 8:24pm, Dec 08)
Last: Meatwad

NewsblogMLB to allow teams to sell local digital streaming rights: Here's what it means
(1 - 8:13pm, Dec 08)
Last: Walt Davis

NewsblogOT - College Football Bowl Spectacular (December 2019 - January 2020)
(4 - 8:06pm, Dec 08)
Last: Walt Davis

NewsblogAn honest look at the Pirates’ frugal free-agent history under Bob Nutting – The Athletic
(12 - 7:55pm, Dec 08)
Last: Bring Me the Head of Alfredo Griffin (Vlad)

NewsblogOT - November* 2019 College Football thread
(602 - 7:51pm, Dec 08)
Last: Red Voodooin

NewsblogNumbers Suggest Mike Trout's Best Days Are Already Behind Him
(38 - 7:49pm, Dec 08)
Last: Jack Sommers

NewsblogCaptain Obvious
(59 - 7:43pm, Dec 08)
Last: Hysterical & Useless

NewsblogOT - NBA Thread, Start of the 2019-2020 Season
(1380 - 7:16pm, Dec 08)
Last: spivey

NewsblogThe Hall of Fame may have a Harold Baines problem
(136 - 5:58pm, Dec 08)
Last: PreservedFish

Page rendered in 0.2967 seconds
46 querie(s) executed