Player Development: Comparing the skating mechanics of Nathan MacKinnon and Charles Hamelin
Using computer vision techniques to analyze skating technique
A few weeks ago, I attended my first in-person speed skating event and it was an awesome experience.
However, after watching a few races, I noticed two interesting habits that speed skaters had when accelerating at the beginning of a race.
First, I noticed the lack of stride extension in the acceleration phase of speed skaters. Their first few strides, at the beginning of a race, were always quite short. For speed skaters, it looked like the number of strides and the footspeed were more important than generating maximum power from each stride.
The second thing I noticed was the ability that these athletes had to keep their upper body stable through their acceleration sequence. This could have interesting applications to hockey, as we will discuss later in this article.
I’m no speed skating expert, but my assumption is that these two technical elements allow skaters to have both speed and stability to navigate the rocky start sequences during which space is limited and positioning is everything.
But before talking about the advantages and disadvantages of these techniques in hockey, I wanted to make sure that these observations checked out against quantifiable data points and if so, by how much.
Computer Vision & Pose Estimation
To ensure these observations were consistent with quantifiable data points, I used several computer vision techniques to be able to pinpoint the movement of different body parts through space and time. My methodology was inspired by this pose estimation project.
And since we are comparing hockey to short track speed skating, I remembered that a few years ago, Nathan MacKinnon had faced Charles Hamelin in an “acceleration race” going from one blue line to the other. Given that MacKinnon is currently one of the best skaters in hockey and Hamelin is one of the best speed skaters of all time, using this race was going to provide us with a good tool to compare hockey and short track speed skating techniques.
Running the computer vision algorithms on the “acceleration race” between MacKinnon and Hamelin yielded the following visual results:
Pose estimation algorithms are by no means perfect. For example, in our clip, there are 10 frames in which Hamelin is not visible, being hidden by MacKinnon in the camera angle. This is an issue as we are not able to estimate the pose of Hamelin for these 10 frames. For purposes of this study, I decided to simulate the movement of Hamelin based on an average of Hamelin’s other frames in a similar posture.
Moreover, pose estimation isn’t perfect even when Hamelin and MacKinnon are fully visible. This yields noise in our estimations of the poses from one frame to the next. As such, to smooth the results for a given frame, I decided to use moving averages on the last 3 frames when calculating the metrics which we will discuss in the next part.
Stride Extension
When trying to analyze stride extension, scouts usually look at the alternation between flexion and extension of the knees through a player’s strides. Therefore, to quantify the magnitude of stride extension, we will calculate the angle of the knees for Hamelin and MacKinnon.
When computing the results through the race, the evolution of knee angles for our 2 racers can be summarized in the plot below:
Visually, we notice in this plot that the blue peaks (representing maximum extension for MacKinnon) are on average higher than the red peaks (maximum extension for Hamelin). This is consistent with our initial intuitive observations, but the real question is by how much?
The maximum stride extensions for MacKinnon average out to 138 degrees, whereas the average maximum stride extension for Hamelin is around 127 degrees. In reality, these extensions should be slightly higher and are understated in our results due to the moving average smoothing applied to the angle calculation.
Consequently, our algorithm identifies 11 strides (+ the start = 12) for Hamelin and 9 strides (+ the start = 10) for MacKinnon, once again, confirming our theory of more strides with shorter extension for speed skaters.
In a hockey context, given the nature of the game, good stride extension allows hockey players to generate maximum power on each stride in the acceleration phase. Moreover, it ensures that the back leg is just under full extension to recover for the next stride making the skating mechanics more fluid.
Upper Body Stability
As for the upper body stability, we can leverage the shoulder angles from our pose estimation algorithms. Instead of solely looking shoulder movement patterns, this time we can analyze shoulder movements in a cumulative manner. This cumulative calculation is simply the absolute difference of shoulder angles from one frame to the next summed up over the entire race for each skater.
Our results can be summarized in the plot below:
Once more, we note that our initial observations were correct. Hamelin’s upper body movements are much more limited than MacKinnon’s shoulder rotations. As such, Hamelin displays much better upper body stability through his acceleration.
Numerically speaking, we are talking about a difference of 420 degrees between the shoulder movements of Hamelin and those of MacKinnon throughout the entire race.
MacKinnon is definitely not the hockey player that moves his shoulders the most when skating. But generally speaking, the upper stability that short track skaters exhibit in the acceleration phase is something that some hockey players can learn from, especially when carrying the puck.
Some hockey players move their shoulders vertically, while others have the tendency of moving their upper body on the horizontal axis when skating. Both of these habits can have negative effects on a player’s ability to efficiently skate with the puck.
This is due to the fact that shoulders are linked to elbows, which are linked to hands, which are holding the stick that is carrying the puck. As a result, inefficient shoulder movement can turn into inefficiencies in stickhandling, hindering, for instance, a player’s ability to be deceptive with the puck.
Therefore, limiting shoulder movement could allow hockey players to be more efficient puck carriers on the transition.
However, at the end of the day, it’s important to keep in mind that the right balance between limited shoulder movement and good stride extension is the key to becoming an explosive and dynamic hockey player like MacKinnon.