PITCHf/x Summit Recap

This past Saturday was the second annual PITCHf/x summit. The summit, put on by Sportvision the creators of the PITCHf/x system, was a chance for interested parties to get together and discuss all things PITCHf/x and HITf/x. I had the pleasure of attending and I wanted to share my highlights. You can go over to Sportvision’s website and download the power points of any of the presentations.

PITCHf/x

Matt Lentzner gave a great talk on arm slots and the movement of fastballs. He showed that the angle of a pitcher’s arm slot is very close to the angle of the vector connecting the (0,0) and (h_mov,v_mov) of a pitcher’s fastball. He showed images of Brad Ziegler‘s and Hideki Okajima‘s arm slots and the movement of their fastballs, which was very compelling. Lentzner continued on how this could be used to make pitch classiciation algorithms better.

Harry Pavlidis talked about the next step in pitch valuation. He, briefly, discussed the current method which he and I currently use, and is used in the pitch valuations here at FanGraphs. It is explained here. The method was introduced by Joe Sheehan at Baseball Analysts. Harry went on to propose two new pitch valuations methods, both defense independent based, one based on batted ball type identification (be they BIS or GameDay or STAT) and the other based on HITf/x data.

Finally we discussed the problem of park to park PITCHf/x differences. The differences are much corrected from last year, but still evident. In addition, we discussed that at even within a season in one park PITCHf/x values can shift. Marv White of Sportvision told us that before every game the system is re-registered to make sure that it is in proper registration. The registration is super sensitive, in PETCO, for example, the weight of additional fans sinks the entire stadium enough to draw the cameras out of registration. Some suggested that Sportvision release the information whenever a system in re-registered so that analysts knew of these events.

HITf/x

Greg Moore and Marv White of Sportvision introduced us to HITf/x. The cameras that capture the pitches for PITCHf/x are on continuously, so they capture images of the ball coming off the bat on balls in play. Thus collecting the HITf/x data was a natural extension for Sportvision. Since the cameras are trained on the space area between the pitcher and catcher they only catch the beginning of a ball in play’s trajectory. Depending on that trajectory they capture between two and fourteen images of the hit ball. From those images Sportvision fits the equations of motion just like they do for a pitch. Since the cameras were set up for the pitch they capture enough images to fit all nine variables in the equations of motion for a pitch (horizontal, vertical and depth position, velocity and acceleration), but they can only fit six for hit balls (horizontal, vertical and depth position and velocity). From this the initial speed of the ball off the bat, vertical and horizontal angle of ball are calculated.

Without acceleration the spin on the ball in play is not known, so the backspin which gives, particularly fly balls, loft, and the side spin which causes hits to slice or curve are not available with the current HITf/x system. Thus the final position and hang time of the hit cannot be accurately calculated. In Alan Nathan‘s presentation he combined the HITf/x data for HRs with Greg Rybarczyk’s final landing distances. With that he could estimate the spin on each HR. He showed that each additional mph in the speed of the ball off the bat added about four feet to the distance of a HR, and showed how the slice of the ball differed coming off the bat of LHBs versus RHBs.

Peter Jensen combined the HITf/x data with the Gameday fielded locations. He looked at plays where the two disagreed on the horizontal angle of ball in play, restricting his attention to balls to the shortstop. Obviously they disagreed on deflected balls, since HITf/x gives the angle of the hit while Gameday where it is eventually fielded. In addition to these deflected balls he found 60 more balls in play to the shortstop where the angle was off by more than 10 degrees. For these the HITf/x angle was more accurate in all but two cases.

BASEBALLf/x

The section on the future was the most exciting. Matt Thomas, Greg Rybarczy and Marv White discussed the future of ball and player tracking. Matt Thomas has tracked fielders at Busch Stadium for a number of games. From the press box using a tripod-mounted consumer level DSLR, laptop and the discipline of photogrammetry he tracked where players started in the field and where they went to field the ball. The furthest a player went to field a ball was around 120 feet by Adam Kennedy running to catch a pop up in foul territory. He also had great information about the probability of a fielder getting to a fly ball based on the distance between where he started and the ball landed.

Rybareck and Marv White both talked about the future of tracking the ball from the pitcher’s hand to the end of the play and all players on the field: runners and fielders. Sportvision is in the research stages of that project and it was recently described in the New York Times. Once this information is available it has the potential revolutionize baseball analysis.

The conference was great. I want to thank the presenters, Alan Nathan who organized the talks and Sportvision for hosting the summit and taking us to a Giants’ game afterwards.



Print This Post



Dave Allen's other baseball work can be found at Baseball Analysts.


Sort by:   newest | oldest | most voted
Big Oil
Guest
Big Oil
7 years 1 month ago

There is going to be a day when we look back and laugh that we once used GB, LD, and FB% to evaluate batted balls.

Thanks for the recap.

Dave Allen
Guest
Dave Allen
7 years 1 month ago

Very true, and based on what I heard it sounds like that day is coming quite soon. Sportvision has the speed of the ball off the bat and vertical angle with the HITf/x data. Those effectively replace GB, LD, and FB. It is not clear at this point when they will start releasing that data, but it shows that the days of GB etc are very numbered.

Samuel Lingle
Guest
Samuel Lingle
7 years 1 month ago

Wouldn’t you need the spin of the ball off the bat to have truly complete data? I’d think it would be fairly similar for each ball of similar trajectory and speed except that the spin of the ball coming into the bat probably affects it coming off, so fastballs and curveballs etc. may differ in some regards when they are batted at equal angles.

Brian
Guest
Brian
7 years 1 month ago

Are we going to have to view that data in graph form, or will a red line be drawn to categorize popups, fly balls, deep line drives, etc.?

Snowball2
Guest
Snowball2
7 years 1 month ago

Thanks for the recap Dave. Im jealous you went and I didnt! Maybe next year!

Eric Seidman
Guest
7 years 1 month ago

Dave forgot to mention that we got to meet! And it was awesome!

Dave Allen
Guest
Dave Allen
7 years 1 month ago

It was great to meet Eric, Harry, Peter, Colin, Max, Matt, Alan, Dan, Greg and lots of other folks I had previously only known through their online writing!

Merkin the Giants Fan
Guest
7 years 1 month ago

I had to leave early for a friends wedding =( and =).
Any word on when more data will be published?

wpDiscuz