Intro to Motion Capture
The best people in sports tech understand the mission. The pursuit has nothing to do with science or software- it has everything to do with improving performance
Every 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 now.
One of the first posts I wrote was on my desire to edit all early biomechanics content for Reboot Motion. My reasoning was simple- I was new to the world and had a lot to learn. By editing videos and picking the best clips to share I would pick up tons of new information.
So, when Reboot Motion doubled down its content efforts by launching Reboot Insiders, I doubled down as well.
So far, Reboot Insiders has hosted two webinars- with Jimmy sitting down with 1) Buddy Clark and Minmin Zhang of Diamond Kinetics to talk about motion capture and 2) Dr. Travis Ficklin and Dr. Robin Lund to go over how physics can be used to drive change.
After each webinar, I did the following:
Posted the full video on our site.
Turned the webinar into podcasts, which can be found on Apple or Spotify.
Cut the best ~1 minute clips and posted them on social media.
My final step is writing a summary on each- sharing highlights and takeaways. While our guests do a great job explaining complicated subjects, I hope an audience member like myself can bring another viewpoint.
First up…
Reboot Insiders: Using Physics to Drive Change
Use Case
Our first Reboot Insiders started with the most repeated takeaway I have heard over the past year: the best people in sports tech understand the mission. The pursuit has nothing to do with science or software- it has everything to do with improving performance.
Just like the best coaches keep their athletes front and center, the best product people keep their customer as the focus, and that is abundantly true with Diamond Kinetics.
—
Sensor Based Motion Capture
As leaders of a sensor first company, Buddy and Minmin had the clearest breakdown of sensors I have heard to date.
Unlike marker based and marker-less systems, which generally have the same goal, sensor based motion capture truly is a different use case.
As the experts from Diamond Kinetics point out, sensors do not understand position because they do not have an anchor. Therefore, sensors cannot capture the data necessary for in-depth biomechanics analyses.
However, sensors do a great job of tracking directional change- and therefore measure velocity and acceleration extremely well. Simply put, sensors are a great solution when we do not care about position…mostly due to their flexibility and ease of use.
IMUs can be placed directly on an athlete-or in DK’s world, on a bat or a ball- with no camera setup. This saves time, saves money, and allows the athlete to focus solely on the movement.
—
Marker Based Motion Capture
Marker based motion capture has been the gold standard for one simple reason- accuracy.
In environments like pro baseball, where any slight tweak could be a million dollar change, sub millimeter precision sounds quite appealing. And for years, marker based systems were the only way to get it.
So how does it work? Markers are placed on an athletes body. As the athlete performs any motion, multiple cameras track each marker’s location frame by frame.
If we know each marker’s precise placement on the body, and we know exactly where the markers are in space for every frame, we can create an accurate skeleton and analyze its movement.
This is all sounds great.
However, there are costs to marker based systems. Most notably, any athlete getting measured actually has to put these balls all over their body.
A good marker based setup involves around 8 cameras, and at least 3 markers on each body part that needs to be a tracked. This makes marker based systems 1) expensive 2) time consuming 3) inconvenient and 4) unnatural.
In the past, these downsides were worth it. Marker based systems were the best way to understand motion. And understanding motion is the obvious first step to improving it.
But with new technology, we can capture data and drive change with less cost to the athlete.
—
Marker-less Technology
Machine Learning
Before understanding marker-less motion capture, it is important to appreciate the technology that makes it better: machine learning.
Not only is ML the driver to making marker-less motion capture possible, it is also the key variable in how effective different systems are.
During the webinar, the group compared ML to giving the computer flashcards.
This analogy was great for me, as I could picture someone giving the computer 1) an image and 2) an answer over and over and over. The computer then teaches itself how to get from A to B.
The biggest driver of effectiveness for any ML algorithm is the flashcards provided. Not only does the quality of the training data have to be high, it also has to include enough variability for the model to work in the wild.
If not, you will end up like Jian-Yang on Silicon Valley.
—
Marker-less Motion Capture
Marker-less motion capture has the same goal as marker based systems: track the athlete’s movement. The only difference is how the two systems build the skeletal model. Marker based systems use cameras to track markers that can be traced back to specific points on the human body. In marker-less systems, the cameras directly track the body…with ML identifying the key points needed to build a skeleton.
Similar to marker based systems, the quality of marker-less motion capture can vary greatly. For marker systems, the key drivers were 1) the number of markers and 2) the accuracy of the people placing them on the athlete’s body. For marker-less systems, there are also two main variables: 1) hardware and 2) software.
Better cameras lead to better results. It is why MLB teams spend six figures plus to outfit their stadiums.
While this is great for the few who can afford it, technology is rapidly improving to welcome everyone else to the party. There are great mobile solutions popping up every day…and the camera we all carry around in our pocket gets better every day.
Aside from hardware, the quality of the machine learning algorithms used matters a ton. This is all done behind the scenes, but the quality of the training data and the engineers doing the work determines the effectives of the system.
(Athletes should keep these drivers in mind. There is huge variability. I would recommend against blind trust in a marker-less system, just like I would be skeptical of writing off the technology if one product does not work well.)
—
Reboot believes marker-less motion capture is the future. First and foremost, sports technology is always about putting the customer first. Less intrusion, less setup costs, and more flexibility is music to an athlete’s ears…and therefore to a coach’s ears.
As cameras (the hardware) and machine learning (the software) continually improve, any advantages of marker based system will dwindle. As that gap closes- and some believe it has already closed- so do the reasons to use a marker based system.