Getting out of the injury zone, part two

(My apologies in getting this article out far too late. Life circumstances and such got in the way.)

In Getting out of the injury zone of Getting out of the injury zone, I discussed a machine learning algorithm that took PITCHf/x variables and attempted to predict what combinations could indicate future injury to pitchers. Today’s article will focus on the scouting part of the equation.

The need for high-speed video

I’ve been a vocal advocate of high-speed videography for a long time now, and strongly believe all clubs should have the capability to record consistent video clips of their pitchers from a variety of angles. Consumer-grade cameras with high-speed capabilities are falling in price, and combined with wireless technologies and synchronization software, videos could be made available to interns over the wire at near real-time speeds. Having a library of high-speed video (example: Mark Appel scouting profile) would be incredibly useful for scouting purposes—and biomechanical analysis methods, too.

What people call “pitching mechanics” needs to be more specifically defined. There are two ways of understanding the movement patterns of pitchers:


Kinematics describes what the body is doing—for example, the humerus (upper arm) is internally rotating (accelerating) at 5000 degrees per second. Kinematics does not tell you the load on the various levers, joints, and pulleys involved in the process. An example of a basic static two-dimensional planar analysis would be:

Mark Appel at SFC

A two-dimensional kinematic analysis can tell you a fair amount of useful information if you have a suitable control object of known size in the shot (we use pitcher height in these images, which isn’t exactly the most reliable indicator)—a ruler or surveyor’s rod in the shot would do the trick. I’ve shown what you can capture in my articles, so if you’re interested in those, go Kyle’s THT Author Archives for plenty of examples.

Two-dimensional video clips with frozen frames can give the coaching staff and front office an insight on how a pitcher looks like at various checkpoints in the delivery (industry-accepted values such as Stride Foot Contact, Maximum External Rotation, Ball Release, etc), which can help if a pitching coach is trying to effect a particular change in mechanics. It’s impossible to evaluate changes—even seemingly large ones!—with the naked eye or at 25-30 frames per second of a regular camcorder. There’s also very actionable data for trainers from side view high-speed video—for example, pitchers who have a significantly internally rotated humerus and a significantly flexed elbow at/near Stride Foot Contact (SFC)…

BJ Ryan at SFC

…will accumulate greater stress in the late-cocking phase of the delivery, as inertial load of the baseball is higher in the Maximum External Rotation (MER) position. While it’s simple to say “Don’t do that,” changing someone’s motor patterns with a snap of the fingers is basically impossible. Extremely fine motor control is a learned behavior over years—decades—of repetitions, and even subtly altering these movement patterns takes a lot of time, effort, and specific instruction geared to making it stick. So, while the person is making changes (or not, as the case may be), trainers can help mitigate the stress caused by this particular flaw. (The details of such methods will be discussed in the third article of this series.)

However, a planar kinematic analysis cannot give you detailed measurements, such as angular velocities of particular body parts. For that, you need a multi-camera setup that combines two or more cameras into a three-dimensional representation of the movement in question. PITCHf/x uses a calibration system much like my biomechanics lab does:

PITCHf/x Calibration Flags Biomechanics Control Object

With those detailed measurements, you can then get the actual joint loads on the body, which is called…


Kinetics describes how the body is doing what it’s doing—or really, what forces are being generated as a result of a particular movement pattern.

A simple way to think about it:

Kinematic Measurements + Measurables = Kinetics

Measurables in this case would be the pitcher’s weight of his humerus, upper arm, shoulder, torso… hmm. How would you know that information? You can’t readily weigh someone’s upper arm or torso without some radical surgery that is likely to leave the patient in bad sorts. Fortunately, some research scientists figured out a way to estimate these values:

Kinetics Estimations

In case you can’t read it, you can find the abstract over at Pubmed (Adjustments to Zatsiorsky-Seluyanov’s segment inertia parameters).

If you’re thinking “That sounds incredibly error-prone,” you are absolutely correct! Consider this: To calculate kinetic data, you are taking a multi-level derivation of positional data. Here’s the stepping logic:

Retroactive Review: Ace
Looking back at some of Justin Verlander's most interesting moments.

-Position: The spatial location of a point in space. Measurement error is the biggest issue here.
-Velocity: The rate of change between two points in space.
-Acceleration: The rate of change of velocity.
-Force: Mass times acceleration.

If what you want to study is the force on a particular joint of the body (assuming a rigid body structure), you have a lot of sources of error there. Especially when you consider the collection method involves manually digitizing each point on the body you want, for each frame of the video, for each video in the camera, for each camera used on a pitcher…


Right. I think you get where this is going.

Is there a better way?

It’s possible to capture data in a laboratory setting using markers on the body, much like American Sports Medicine Institute (ASMI) does:


By using highly reflective markers—and some interesting math—you can create a representation using multiple cameras to film the subject in motion. While measurement error is no longer an issue, two problems arise. The first is minor, but does affect consistency—how do you know the researchers are applying the markers in the same location every time? While there are specific instructions for placement (place on the distal elbow at the olecranon process, for example), there are bound to be errors here. The second is a major issue that gets glossed over: Release velocities are NOT identical in the lab and in a game situation! The reasons for this are varied, but the largest is pretty obviously the fact you are covered in sticky markers, throwing off a turf mound, into a net, in a lab.

Let’s investigate that problem a bit further: We know that velocity is not the same in the lab as it is in the games. Can we assume that the drop in velocity is constant between pitchers? And furthermore, do a pitcher’s mechanics meaningfully change when they throw off a turf mound under analysis as compared to a game situation when they are throwing much harder? And do the angular velocities, accelerations, and forces all scale predictably between their lab results and game results?

Even if you could get a precise calculation of joint loads, you wouldn’t know the tension on the ulnar collateral ligament (UCL), the labrum, the distal elbow, or the various rotator cuff structures—all common areas of injury. Due to anatomical variations, muscle fitness, and combinations of movement patterns, it’s impossible to say “The UCL is under X Newtons of force at this point in the delivery.”

Despondency with the methods

OK, OK, so it all sounds bad – there’s no real way to get a true answer. However, that doesn’t invalidate the fact that data collection can be useful. A true three-dimensional setup may not measure specific numbers X, Y, and Z, but it could very well give us answers anyway! It’s entirely possible that collecting live kinematic game data and analyzing it using statistical methods (regression, machine learning, data mining, etc) could yield very useful results. Given the fact that consumer-grade cameras are inexpensive and a setup as described in this article can be built by laypersons (yours truly, for example), it could absolutely grant an edge in scouting to teams—where such edges continue to disappear.

Meanwhile, advances in wearable technology will continue to improve—at Driveline Biomechanics Research, we’ve partnered with MotionShadow to trial their outstanding non-intrusive kinematic collection system, and we’re hard at work on a WiiMote-based solution that captures pitching arm kinematics (specifically interested in the relationship between forearm pronation timing vs. maximum internal rotation velocity). These technologies could be useful for low-cost fast-track analyses, provided that release velocity stayed close to competition release velocity and pitching mechanics were not significantly altered.

By combining the data collected from a high-speed video setup, wearable computing data, and PITCHf/x data, you could start to explore some seriously interesting frontiers for both performance enhancement and injury prevention. In part three of this series, we’ll talk a bit more about the details of the applications of this technology.

Print This Post
Kyle owns Driveline Baseball and Driveline Biomechanics Research, and has authored The Dynamic Pitcher, a comprehensive book and video set dedicated to developing elite youth baseball pitchers. He is also a consultant for an MLB team and a major Division-I college program. Follow him on Twitter @drivelinebases or email him here.

Comments are closed.