from Amazon
Tuesday, January 29, 2008
Marv, CTO of Sportvision, answering some PITCHf/x questions
Big thanks to dedicated reader Peter Jensen for asking and transcribing everything. Long-time PITCHf/x followers may find it convenient to skip over bullet point #1, and just come back to it at the end of the dialogue:
1. From Peter Jensen (fan):
Marv - Thank you very much for taking the time to discuss the capabilities of the pitch f/x system with me. I am part of a group of amateur baseball analysts that are very impressed with the data that is being generated by pitch f/x as it is being reported through mlb.com’s Enhanced Gameday. When the system is implemented in all the major league parks we will have a tremendous amount of additional data on pitch speeds and movement that we feel will yield new insights on how baseball is played.
But of course we want more. We feel that the most important information that is unavailable to us from any source is precise data on the hit ball. From what we think we know about the technical aspects of how pitch f/x produces the information on pitched balls, it seems that it might be possible for pitch f/x to give us much of the information that we desire on hit balls with little or no modification of either the hardware or software. Specifically, we are looking for the speed of the ball as it leaves the bat and the initial horizontal and vertical vectors of its flight. Depending on how long the ball can be tracked with the camera set up that you now have for pitching, it would also be desirable to have values for departure from its initial straight line path. This would be analogous to the values that you generate to represent the curvature of the pitched ball and would allow us to calculate the horizontal and vertical magnus forces on the hit ball.
Our reasoning has been that the information we are seeking is essentially the same information that your existing system provides on pitches, but in the opposite direction. We understand that your existing camera set up only covers a limited area, but the speed off the bat and initial vectors should be able to be calculated from three or four video frames. If we are incorrect in our assumptions about how your system works, it would help us to know exactly why this information can’t be generated from pitch f/x.
Having this information on hit balls will be useful for a number of cutting edge analyses. Hit ball speed off the bat is the best indication of a well hit ball. Estimates of high hit ball speeds have a high correlation with base hits. Having accurate calculated hit ball speed information is crucial for evaluating how well both hitters and pitchers are performing. It would also be extremely important in long range studies of player aging and recovery from injuries. Both hit ball speeds and trajectories will fill in gaps of information that is needed for better studies of fielder’s abilities, an area that is only beginning to be studied in depth. The pitch f/x system is the only source that we have for this information.
In our discussions with Cory Schwartz about including batted ball information in Enhanced Gameday we have gotten the impression that he believes that we are asking for something that will require a lot of additional work and added costs. If we can get a technical opinion from you that it can be easily implemented on the existing pitch f/x system, it would improve our chances of convincing MLB to include this information in Enhanced Gameday as soon as possible. While our selfish motivation is the research opportunities that having accurate batted ball information would provide, we also believe that the many of Gameday’s more casual fans would also benefit and enjoy this information. Just as a fan appreciates how difficult it is to hit a 100 mile an hour pitched ball, we feel that he or she will appreciate how difficult it is to field a 120 mile per hour hit ball.
Thank you for your help.
2. From Marv, Chief Technology Officer of Sportvision:
Thank you for your interest and diligence in analyzing data. I have taken the liberty of copying several people on this email including Cory Schwartz, whom you know of course.
First of all, Cory is right that computing the velocity (speed and direction) of the ball off the bat does require more work. I do believe the work is entirely software development and deployment; no new hardware required. I am confident we will be able to automatically find the hit ball in a few frames of video after the hit. It would be many frames if the ball were hit right back to the pitcher, but usually the ball leaves the field of view of our tracking cameras quickly.
My confidence that it is possible stems from looking at numerous video clips myself but more importantly from the work Alan Nathan did to compute the very values you hope to get (I don’t know if he published his work yet).
Can we do that work? I’m confident we can. Can we make it high enough priority to do it soon? I’m not so sure about that, but it is definitely of interest.
So, Peter, please regard my reply as indicating strong interest and high level of appreciation for your interest as well. Beyond that, it is also part of the discussion for setting priorities that MLBAM and Sportvision carry on as a matter of course. Your suggestions and enthusiasm are welcome and will certainly influence our choices.
Thanks again
3. From Peter:
I also want to make sure that I understand the software issues correctly. During our phone conversation my understanding was that the existing software could make the hit ball speed and horizontal and vertical angle calculations without out modification or with only trivial modifications. The area in the software that might require more intensive effort was having the software automatically correctly identify a hit ball situation and then discriminate the correct location of the hit ball as it leaves the bat from other extraneous movement in the subsequent frames. Is this a correct assessment of the problems? I was also under the impression that the amount of work necessary to solve these potential problems could not be anticipated until tests were run on enough actual game footage that would contain a variety of balls hit at different angles plus non hit ball situations. Is this also correct?
4. From Marv:
I can’t go into the details of our proprietary tracking software, but I can tell you that we now track the ball flying from the pitcher towards the batter and purposely ignore anything going in a different general direction. Tracking the ball off the bat will require recognizing objects travelling in nearly any direction. That isn’t difficult, but it is a change and requires careful testing and deployment, a typical requirement for any software in general use.
5. From Peter:
[We’d] like to know if you would be interested in participating in a question and answer session sometime in the future at your convenience on the technical aspects of the Pitch f/x system in general.
6. From Marv:
I would be happy to join a chat session.
Nicely done, Peter. I think the value of HitFx is obvious, and it will float to the top of their list once they give it some consideration…
One thing to note, while fly ball analysis will be possible from initial “launch” data, using something like Hit Tracker, launch data for ground balls is not going to be actionable - the inelastic collisions with the ground are not likely to ever be “modelable” to the point where a system can take only the first few feet of a grounder and predict accurately where and when it will intersect with the fielders. We’re still going to need observation data on the ball as it nears the fielders.
We’re also going to need positioning data on the fielders. I have some data from Hunter and Jones, pairs of (virtually) identically struck grounders, as measured by time required to reach a certain spot on the infield, that highlight the importance of initial positioning; one ball rolls straight into the glove of a fielder who doesn’t even need to move, while another rolls into the outfield untouched. The same exact ball path and speed along it, I repeat. When you see the examples, I expect you’ll agree that knowing the ball path alone (which HitFx will not be able to provide, I repeat) is not going to satisfy us, we are going to instantly want fielder positioning.
When I have the data in presentable format I’ll post it here.