Diagnosing and Modeling SSW pitches: Post 58

I’ve carried on about this, but if you throw a Seam Shifted Wake pitch, there is currently no way to know that it moved differently than any other pitch. I’ve known for some time that this was an obstacle to our work. I’ve been told many times that these effects aren’t real because Trackman and Rapsodo do not show them happening.

We may see them with our eyes, and the really special ones look unbelievable, but we always have to ask ourself, “did the camera fool me here?” It’s a good question because different camera angles have a huge effect on our perception of pitches.

This spring, before the virus struck, I was invited to the Blue Jays spring training camp by Matt Buschman (thanks again). While there, I was able to also meet with the Yankees and the Rays. It occurred to me during that trip that unless people can measure these things, it’s never going anywhere. Pitching is iterative, and if you don’t know it was good, how can you iterate?

If one knows how the ball moves in 3-D versus time, SSW effects should be apparent. As an aside, Trackman measures 3D motion, but cannot measure the axis of the ball (note that it is common for 3rd parties to add information on the axis gleaned from the ball’s movement, but if a SSW is happening, this information is incorrect). Additionally, it outputs the results as a “9-parameter model” that I believe wipes out some of the effects I am interested in. Their older systems may have not been sufficiently precise to warrant any other way of outputting data, but I am convinced their newer ones are. And the MLB Hawkeye system that is coming whenever baseball gets back to normal may too.

But why wait? I believe it is possible to measure SSW effects using only a single camera. The camera must be set up somewhat carefully, but this requires nothing more than a level, a tape measure, and 60.5′ of flat ground. I am hopeful that the video could even come from your phone.

I have been working with two pitchers on this and they’ve both been incredibly helpful. Thanks to Connor Нinchliffe, who is in the Phillies organization, and to Jared Hughes, who has been a major league reliever for almost a decade. They have each been working on a setup and sending me their videos.

The video below is a 4-seamer from Jared’s most recent setup. I asked him to throw this because I wanted a pitch that did not have SSW effects as a baseline, and 4-Seam fastballs generally do not (foreshadowing).

4-Seam Fastball

Jarod’s bread and butter pitch is a sinker. He has a low arm slot and throws the fastball at a tilt of about 2:10-2:20 (similar to the 4-Seam), so it does not get much ride from Magnus force like fastballs from a higher arm slot. Still, the vertical movement on this pitch is pretty amazing.

Sinker

He has overlaid the 2 pitches here:

Jared jumped on one of my Zoom calls a bit ago and sent me a video of one of these from a game and I knew immediately that this guy had it going on.

The orientation of his sinker is just about perfect. I would not know how to improve it. And, Trackman knows it is good. The histogram below is A_z, or vertical acceleration, from baseballsavant.com of his sinkers normalized by acceleration due to gravity. (Some have warned me not to use this source, but I have consistently found that, while the data is presented differently, I would make the same conclusion if I used Brooks). So if it is to the left of negative 1, it’s being pushed down, it’s not falling. His tilt on this pitch generates a small amount of positive A_z, so an overall value below 1g means the seams are doing some work. I have not found another fastball like this in the majors other than submarine guys, and they get it from Magnus force. (Readers, let me know if you know of one!)

Vertical acceleration of Hughes’s sinker for the last several seasons

The question that I set out to answer is: can one find the X-Y-Z trajectory of a pitched ball from a video like this? And if I had that, could I detect a SSW pitch?

I should mention that the setup here is important. He followed the recommendations of my earlier post. Jared knows how high the camera is off the ground and has tried to make that match his release. Additionally, he knows how far behind the rubber the camera is, and how far in front his release is. Lastly, he provided me with the size of the 9-pocket behind home plate.

I have written a Matlab script that loads in a video like the ones above, removes the background, and finds the ball at each frame. Knowing the release velocity of the pitch, we can estimate how much it moves toward the plate between each frame (while considering drag), and correct the vertical position by accounting for parallax based on the known distance between the plate, the rubber and the camera.

These trajectories can be used to estimate the vertical release angle, which turns out to be the same for these two pitches (-1.57º), and with this information plus Rapsodo metrics, I can simulate these pitches using Andrew Smith’s UMBA 2.1 code. The results are really interesting. First, ignoring the seams, we get this.

Trajectory of 2 pitches extracted from the video (symbols) and from UMBA 2.1 (curves) with no seam model.

As far as a Magnus model (which is what UMBA is with no seam model), the two pitches are nearly the same. There is a small difference in their tilt leading to the small difference in z at the plate.

In order to model the seams, we need the orientation of the ball relative to its axis for both pitches. I do not know of a tool to estimate orientation based on video (someone should make one), but I can do it iteratively with the help of Trip Somer’s spin simulator. Based on the videos (which I have in HD), I estimated this for the sinker:

Sinker orientation

It’s easy to see the seam on the top front of the ball about 3/4 of the time (note this is the pitcher’s view).

The 4-seam looks like this:

4-Seam orientation

The “Top” orientation is z in UMBA and the Front is y. Entering this information and turning on the seam model gives this result.

Trajectory of 2 pitches extracted from the video (symbols) and from UMBA 2.1 (curves) with the seam model.

I expected UMBA to predict more sink on the sinker. That’s what I am here for. BUT, I was pretty shocked to see it predict that the 4-Seam has additional vertical break. I had to go back and look carefully at the 4-Seam orientation again. There are seams near the top and bottom of the ball about 1/2 the time. From the model output as well as the measurement, it seems that the seams on the bottom are winning. But why?

Note that the seams on the bottom are consistently just forward of the hemisphere plane. The seams on the back are just behind it. They can both hasten boundary layer separation, but separation on the front of the ball has a much larger effect.

In other words, his 4-Seam has carry due to seams. And he knows that it plays well in the top of the zone. Here’s an anecdote with everyone’s favorite fighting second baseman.

In conclusion, here is what I think I have learned:

  • Seams matter (again)
  • Jared Hughes’s sinker uses seams to create sink
  • And his 4-Seamer uses seams to create ride
  • I can measure the trajectory of a pitch with a single camera carefully placed behind the pitcher

On that last point, I am very interested in making this more accessible, and I think a phone app may be the way to go. I tested this out last night with my iPhone, and I got some promising results, without even using slowmo!

Related Post

Leave a Reply

Your email address will not be published. Required fields are marked *