The Pitfalls of 2D “Biomechanical” Analysis
Everyone wants to do what they can to prevent baseball injuries. With the increase in different technologies (wearable technology and high-speed cameras) and decreases in prices, there has been a shift in what coaches and players can do to examine their performances.
Of course, as helpful as this technology can be, it doesn’t come without some people taking what they think are conclusions significantly too far: specifically, when coaches are trying to use “biomechanical” analysis.
The truth is that most “biomechanical” analysis is increasingly flawed—not only in the process used to collect it but it the conclusions reached from such analysis. Most conclusions from high-speed video or pictures are taken increasingly further than they should be.
So, let’s try to answer this question: Can you take biomechanical measurements from pictures of slow-motion video?
First a Recap: 2D versus 3D
It may seem too simple, but 2D is essentially looking at something in one plane, thus a two-dimensional space. This means you are looking at something with an X and Y plane, you can look side to side, up and down, but you can’t measure depth or rotation—this is important. Pictures and video all take place in 2D.
On the other hand, 3D adds depth. This means you can look at the X and Y plane and also the Z plane. So, you can look side to side, up and down, and also see depth. In order to have mechanics measured in 3D, you need to have at minimum 2 cameras recording the athlete simultaneously.
Think of the difference between a circle and a sphere.
At Driveline we have used two camera angles to do biomechanical analysis in the past using the Direct Linear Transformation. Using two cameras (and a lot of math) you can then mark parts of the body and get biomechanical measurements.
An upgrade to this system would be a marker-based biomechanics lab, which we now have. Markers are placed on specific body parts so the cameras can calculate the position of the markers and, therefore, how the body moves.
Just understanding these simple differences between 2D and 3D can tell you that 99% of the “biomechanical analysis” that you can find on the internet aren’t actually biomechanical analysis. Why? Because they are pictures of videos taken from one view point, making it a 2D analysis.
Baseball requires movement in all three planes of motion: sagittal, frontal, and transverse, which mean you move side to side, up and down, and rotate respectively. Pitching and hitting are highly rotational; you can’t get a measure of rotation from any view from one camera.
A true biomechanical analysis means you can measuring athletes very specifically in order to get two things: kinematics and kinetics.
Kinematics is the motion of points without considering the forces that cause the motion, like hip or torso rotational velocities.
Kinetics is looking at motion and its causes, its forces and torques, an example would be elbow and shoulder torques. Sometimes, this is more commonly referred too amongst coaches or players as the “stress” on an athlete’s joint.
Understanding a biomechanical analysis means that you are getting kinematics and kinetics, giving us one answer to our question: “biomechanical analysis” using pictures and videos isn’t truly possible. You can’t get kinematic or kinetic measurements in two-dimensions, like pictures or video.
Of course, they are used very often, and we can still take a closer look at how they are used in order to look more specifically at why many of these analyses are answering questions in the wrong way.
This is a broad look at why 2D “biomechanical” analyses don’t hold water. But it is important to realize that video is still very useful for athletes and coaches you just can’t get biomechanical measurements from it.
People should understand the specific limitations of video, what we can do is explain further why many are inaccurate, so you can get a better understanding of what you can and can’t take from any analysis you see online.
Parallax Error: What It Is and Why It Matters
First off, we have to talk about parallax error. This is something many don’t know about, but it is a common flaw ignored by a lot of analysis.
Simply put, parallax error means that if you are trying to measure the same thing from different angles, there will be different measurements simply because of the different camera angles.
The object in the pictures above hasn’t changed at all. But the measurements of the object change drastically.
If you try to get mechanical analysis based off of still pictures of slow-motion video, you will get different measurements simply because of the camera angle. It’s very common to draw lines, circles, arrows, and even angles on pictures and video. But if they are not taken from exactly the same position, more specifically the same angle and same distance, then the measurements are going to have a large margin of error.
Bob Keyes explains this simply when he is talking about comparing 2D to 3D analysis.
This makes these analyses not reliable, you can’t accurately compare two-different videos or pictures that were taken from different angles.
In order for comparisons to be reliable, the video needs to be taken from the same spot and same distance to what it is being compared to. If you are trying to analyze video of your athletes then mark a specific spot on the floor where a camera will be put every time.
This likely won’t stop people from comparing a pro to an amateur hitter using two-different angles, but it is something people should know. There is a large margin of error in measurement that can’t be accounted for (because it’s a 2D picture).
With the growing amount of cameras, it is interesting to look at different pitchers and hitters to see differences, but that’s about all you can confidently say they are: interesting. You can’t get a measurement of torque by looking at a picture; it looks like you can from a video but you can’t. Pictures and video should be used as a filter to check what you see with some sort of actual biomechanical measurement.
This isn’t currently that feasible for many, but things are changing, and there will be growing access to easier biomechanical measurements in the future.
Still Pictures: Sample Size and Confirmation Bias
The way that people use still pictures and slow-motion video to “yay or nay” good mechanics tends to focus around one main idea: a picture or video is representative of what a pitcher’s mechanics will always be.
For a number of reasons, this is false.
First off, it’s understood that pitchers don’t repeat their mechanics exactly the same for each throw. This can be understood by looking at the release points of a pitcher. You can see that pitchers produce a range of mechanics, not an identical carbon copy each time.
Second, keeping in mind what we just said, know that if you take a picture of a starting MLB pitcher from a game he throws 100 pitches in, you are trying to take conclusive evidence from a 1/100 sample. This doesn’t include the warm-ups between games, of course, which if you are using a picture found on the internet, you don’t know always know if the picture was taken from a live pitch or a warm-up. If he threw all 8 pitches to warm up between innings and threw 100 pitches in 6 innings, now you are trying to make very strong conclusions from a 1/148 sample. Multiply that by, say, 28 starts, and now you’re at, well, you get the idea.
Many coaches often forget that still pictures that they can find online are often chosen because they look good and not because of any sort of scientific reason.
This leaves still pictures as the perfect choice to be chosen by confirmation bias. Pictures are chosen because they fit what someone wants to see. There are only a few pictures to choose from, and the one that is picked for analysis is the one that proves the point a coach is trying to make. This isn’t a good way to make decisions.
Every one of these pictures can be use to support some sort of pitching mechanics theory.
Many people want to build a theory on how to build a “perfect pitcher” by funneling information that points to one conclusion. The problem with this is that a good theory gets boxed in by information that supports it and information that doesn’t support it. Good theories get stronger by trying to prove them wrong, not by being proved right.
Also, narrowing down pitching to a number of “checkpoints” is simply too simplistic. It’s more important to see how things relate to one another.
While there are checkpoints that studies have shown to be important, you still can’t get as accurate of a measure from a still picture as you can from true biomechanical analysis.
Let’s take the angle of when an arm is at foot plate. This picture from ASMI’s lab lets you estimate the degree that the arm is at:
But what you can’t see is the angle that the torso is to the camera, because it isn’t perpendicular, or what angle the humerus is to the camera. It comes back to the fact that you can’t get a measure of depth from a 2D picture like you see above.
So can you measure the angle the arm is at from the side angle? Sure, but you have to keep in mind that there are limitations to that measurement (can’t measure depth), so it isn’t going to be accurate.
Every analysis that you have seen that does not have a camera lined up directly 90 degrees to the pitcher is measured inaccurately. Even then, there are only a handful of relevant measures you can get.
The only accurate measurements that you can get from a 2D analysis are going to be ones that are parallel to the camera. This only truly leaves you with the angle of the front leg and angle of the torso. But even those will have slight margins of error depending on how pitcher’s stride.
While we’ve largely talked about pitching to this point, the same rules apply to hitting. The added difficulty in analyzing hitting is that hitting is reactionary. Each swing is completed based on vision information, meaning the pitch a pitcher is throwing and what the hitter is expecting are going to affect hitting mechanics.
High Speed Video Analysis: Looks Better, Same Problems
There is no doubt that high-speed video can be helpful and interesting to both athletes and coaches. But video runs into the same issues as still pictures—you can’t get torque measurements from one camera video.
The biggest issue with video is the lack of ability to measure rotation.
No matter how many lines, circles, or arrows I put on this video, it won’t be able to read the rotation of his arms. A one camera view video can’t tell you how fast your hips, torso, or arms are rotating.
This is the obvious issue in baseball because it is highly rotational and involves all three planes of motion.
This is why the best-case scenario for getting data is the same as for still pictures: body parts that are going to be parallel to the camera.
Take an example from Kinovea’s website, the angle of a cyclist’s leg when he’s training. This works because the camera is set up directly to his side and his leg is parallel to the camera, which minimizes the chance for error.
Analysis from a biomechanics lab and from a camera analysis also differ in how positions are picked. There are specific measurements that labs are able to get in order to ensure accuracy. So if you are reading an article that talks about body positions at front foot stride, there are specific parameters that measurements of the body can be taken at when the front foot hits the ground. While if you want a video analysis, what qualifies at “foot hitting the ground” usually depends.
Lastly, biomechanics analysis need to know specific body markers to get measurements. This can’t be done while an athlete is in uniform and is why markerless labs get different measurements than marker-based labs.
But the same issue can be said for pictures and 2D video. If someone is trying to measure the angle of a joint, is the point they are trying to measure from accurate?
This is the same video stopped at the same point with different measurements based off of different points.
Via pictures or video, coaches and players need to be aware that some things they think players are getting better or worse at, could be different because of measurement error.
These are all things that coaches and players should be aware of.
There are some serious flaws in believing that you can use pictures and slow-motion video for biomechanical analysis. The core of the issue is that you can’t get actual biomechanical measurements from a 2D view point; you need to measure in three dimensions.
This likely won’t stop people from continuing to do these analyses, because they are relatively simple to do. What it should do is make you realize that the conclusions you can take from those analyses must be taken with a large grain of salt. Many of the measurements are riddled with large margins of error that are often ignored.
Video and pictures can still be useful, there is a difference between looking at a players movement and trying to measure biomechanics. If you still use video to look at players mechanics, then you need to make sure the camera is at the same angle and distance every time. Comparing videos over time can give a better grasp of how an athlete’s movement has changed, but you just can’t tell how forces have changed.
This article was written by Research Associate Michael O’Connell
Neal Walther -
If recording, when you say using the same location every time, I assume you mean using a physical reference to center camera? Such as 3 feet in front of rubber, at height of pitchers hip, or center at batters belt buckle, at waist high, etc.? Should there be a constant and variable?
What are your best practice recommendations of physical camera focal/ center points for a 2 camera system recording simultaniously, for mound and for plate?
MICHAEL O'CONNELL -
Same location referred to the placement of the camera in the gym or on the field. So if you are recording from a baseball field the camera be set up in the same post at each stadium/park. For a gym you can mark a piece of tape on the floor at a certain distance from the mound/batters box that the camera should be used from.
Videos from different camera locations aren’t going to be able to account for parallax error.
We used a pvc cube as a focal point for our 2 camera system (which you can see in the Straily gif above) to center the cameras. Then just removed the cube and had the athlete throw.
Bill Masullo -
Nicely done, Michael! Thanks for the advice. I typically use the video for a sense of tempo and timing and have been aware of issues (never to this degree) in the past. Limited use for “fixing” more so for modeling and reacting to movements unfelt or unsure.
Ricky Norton -
If there is 30 yrs of collecting live game video, they definitely didnt have markers, so that would also put in question the accuracy of the claimed angles of each hitter/pitcher. Markers are the gold standard, along with 3D cameras, correct? PS, Love coach Pullins, he is my guy.
MICHAEL O'CONNELL -
Right, every system has some limitations but markers are the gold standard right now.
In an ideal world you could do a marker set up and a non-marker set up at the same time. Do analysis on both and then compare the numbers and see what adjustments would need to be made from there.