Friar Watch

I found some interesting information about the MLB Gameday pitch tracking. It’s provided by Sportvision, the same company that does the 1st down marker for NFL games. The same system powers the Carls Jr Pitch Tracker, as well as the Fox system that shows the flight path. Note that this is not the same system as Questec, which is used solely for umpire “training”.

Over at the Sons of Sam Horn message board, poster AlanNathan has some insight into the Sportvision system:

Sportvision uses three cameras altogether, two for tracking the pitch and a centerfield camera for establishing the upper and lower bounds of the strike zone for each batter. The tracking cameras are typically located high up behind home plate and high up down the first base line. Together these two cameras (operating at 60 frames/s) track the pitch and recreate the coordinates in real time. A very clever algorithm is used to “find” the ball in each frame (the precise algorithm is proprietary). Each camera finds the pixel location of the ball in each frame. The cameras have already been calibrated so that the pixel information is translated quickly into coordinate information. The coordinates of the ball in each frame are fitted to a smooth curve, and that curve is extrapolated to home plate to determine the location of the ball as it passes home plate. Sometimes the cameras can actually see the ball as it passes home plate but not always, especially for left-handed batters. All of this is done in real time.

How accurate is the system. If I had lots of time, I could probably figure that out. The Sportvision people have been very nice about sending me compete pitch data from games. If I had the time and patience, I could play the game back (downloaded from mlb.com) pitch by pitch, comparing the computer logs with what I see and what the umpire calls. I have done that a little bit and mostly I would say that what Sportvision reports agrees with what I see. Occasionally the system really messes up badly, worse than any umpire would mess up.

Based on my limited observations I would agree that it seems pretty accurate but is prone to being completely off base.

MLB.com has an interesting article about the Sportvision system as well. They note that it provides more accurate pitch speeds than radar, which is prone to a number of variables:

One thing all fans should know is that the speed represented by this pitch tracking system has no reliance on radar guns. They are likely to gradually become a thing of the past in baseball. Remember when Tigers reliever Joel Zumaya was hitting 103 mph during that pivotal Game 2 of the American League Division Series at Yankee Stadium? Many people in the media were qualifying it as the possible result of a “slow gun,” the term itself a dying relic of baseball conversation.

Zumaya’s final three strikes thrown to Alex Rodriguez in the eighth inning of that game, according to the Enhanced Gameday showing speed at release and then speed crossing home plate, were as follows:

102.5/93.4
102.2/91.9
100.2/90.7

How did we know that? How did we know the speeds of Glavine’s momentous pitch to Pujols here on Tuesday night? It’s not about radar guns. It’s about video at 30 frames per second, triangulation involving three ballpark cameras, and software on three computers in the corner of a TV truck outside the stadium’s loading docks.

“What we’re tracking is very accurate pitch trajectory and speed, and of course location in the strike zone,” Shaffer said. “Radar guns do a couple things. Sometimes average speed, sometimes after it crosses the plate, sometimes as it leaves the hand. It’s not just focused on the ball.

“We take a series of high-speed photographs, 30 frames per second, as it moves from mound to plate. At each frame, we can identify the location of the ball. So if we take the time of that frame, we can calculate speed. [Radar guns] are the old days. It doesn’t get any more accurate than this.”

What about the radar readings shown on stadium scoreboards? It’s commonly accepted that teams will tinker with the readings to make the hometown players look better but the real reason for unusual readings is more likely a result of gun placement. This article from the Sportvision website (reprinted from the NY Times)elaborates on the Zumaya readings:

But on Tuesday night during Game 1 of the American League Championship Series, Tigers reliever Joel Zumaya unleashed two fastballs in the eighth inning that were measured at 103 m.p.h., and a third that was clocked at 102.

Or maybe he didn’t. The triple-digit readings were cited by Fox Sports, with its FoxTrax system, which uses a system of three cameras and computers to determine the flight and break of a ball and its speed.

At ESPN Radio, Jon Miller and Joe Morgan spoke of the same Zumaya fastballs, coming in 3 m.p.h. slower than FoxTrax said they did. ESPN’s readings came from a Jugs radar gun positioned behind the plate at McAfee Coliseum in Oakland. The readings were posted on the scoreboard. In fact, all pitches were generally 3 m.p.h. faster on Fox.

“We look at the stadium speed, but if there’s a disparity with what our colleagues have on TV, we mention it,” said John Martin, the executive producer of remote broadcasts for ESPN Radio. “If the disparity is 93 to 96, we don’t mention it, but when somebody hits the stadium gun at 100 and Fox has it at 103, then it becomes a discussion point.”

A few weeks ago I watched the Daisuke Matsuzaka - Felix Hernandez matchup at Fenway and thought the speed readings on Fox (or was it ESPN?) were juiced. They routinely had King Felix at 98-99mph and Dice K at 96mph but they didn’t look that fast to me. At the time I assumed Fox(ESPN?) used the stadium gun for their readings. Knowing they use Sportvision it makes more sense, I would bet the stadium gun showed 2-3mph slower.

Another interesting aspect of this is pitches are routinely lose about 10% of their velocity by the time they cross home plate. Perhaps it’s more accurate to say Zumaya was throwing at 97 or 98 mph, the average of the two measurements, rather than 103 mph.