Building Baseball's Best Single Camera Motion Capture
Single camera motion capture is one of new releases we are most excited about...and we think we're on our way to being "baseball's best".
Every other week I write an email discussing what I learn launching and growing Reboot Motion. If you would like to receive it directly in your inbox, subscribe below.
“Baseball’s best single camera motion capture”.
Reboot Motion is doing a lot of new things in 2023, but that may be the one we are most excited about.
In 2022, we offered our partners a unique, competitive advantage- the ability to analyze every pitch and every swing from every game their pro team plays.
It is how we ensured our partners had the best information, and how we ourselves processed over two million movements per month.
But MLB organizations reach far beyond what takes place in a big league park. The best ones take a wholistic view throughout their system, streamlining strategy and player development.
As a result, we need to offer the ability to go beyond the pro ballpark. We need to offer similar tools:
At minor league affiliates,
During spring training,
While throwing bullpens, and
On the road scouting tomorrow’s talent.
In other words, to fully serve the entire organization, we need to add optionality to our biomechanics train.
Previously, we told teams to send us mocap data and get back skeletal data, movement data, and reports.
Now, they’ll have another option:
Defining Baseball’s Best
Turning raw video from a single camera into motion capture data is good. But pro teams demand more. They demand the best. And they demand evidence.
The Work
Recently, we analyzed the accuracy and repeatability of our single camera motion capture technology to a KinaTrax lab motion capture setup, which we considered our ground truth. We have high conviction in the quality of KinaTrax data, and it is widely accepted by many MLB organizations.
We tested our single camera solution with the simultaneous capture of a single subject- a male RH pitcher with MLB experience- throwing a bullpen session at ~90 MPH on an indoor mound using both 1) a KinaTrax lab setup and 2) slow motion 240 fps iPhone 12 video captured in landscape mode from the pitcher’s open side.
The Results
Our results are in the below pdf. I encourage anyone interested to stop here and read the white paper, written by our CEO, Jimmy Buffi.
For those who want the TLDR, here we go:
Reboot Motion’s single camera motion capture delivers accurate enough data for MLB organization’s to use when a high end motion capture system is not available.
Reboot Motion’s single camera motion capture accuracy compares favorably to the current state of the art single camera motion capture accuracy, which to date has been PitchAI.
Specifically, the number one data point we look to is the correlation of the throwing arm joint angles calculated using Reboot Motion’s methodology with KinaTrax as the ground truth. Our study found an average r^2 of .89 with a RMSE of 11 degrees for the pitching elbow. This was a stronger relationship than was found in the PitchAI study.
Finally, in addition to having strong correlations, Reboot’s throwing arm angles were more repeatable- at around 3 degrees.
(It is worth noting that PitchAI’s validation study did show higher r^2 values for some body parts, but all were quite close. We place high value in the throwing arm, but still believe PitchAI is a reasonable single camera motion capture option.)
The Objection
For anyone poking holes in our data, they likely would start by the lack of variability. We understand there is a need for a wide variety of pitchers, specifically with different deliveries. Any future work will touch on this current shortcoming.
However, it is worth noting our subject pitched with a 3/4 arm slot while previous studies focused on pitchers with overhead arm slots .
3/4 arm slots are a more difficult capture with a single camera from the open side because the arm is not rotating in a similar vertical plane to the camera lens.
Future work- by us or others- should look at a variety of arm slots to accurately showcase the abilities and shortcomings of single camera motion capture, especially as it relates to its use in Major League Baseball.
Additionally, some could question our decision to rely on a markerless system as “ground truth”. While there are academic arguments to compare our results to a marker based system, we prefer markerless for two reasons: 1) it is less intrusive for the athlete and therefore puts them in a more normal environment and 2) it is what MLB teams actually use.
The Takeaway
Our MLB partners have an extremely high bar.
However, organizations have many interested parties, each with different needs and pain points…and therefore different bars.
Biomechanists and data scientists need to deeply understand the data they look at. This validation is just one step in constantly gaining their trust.
Pro coaches get to enjoy the best markerless data possible- both 1) in-game and 2) from pitching labs. For them, the bar is equally high because it can be.
But for scouts on the road? For a front office looking to evaluate players across their organization and far beyond it? For them, the questions are simpler:
Can I trust the data?
Can it improve our scouting?
Can it aid in player development?
And for them, we are proud to answer Yes, Yes, and Yes.