Article
Estimation of Mechanical Power Output Employing Deep
Learning on Inertial Measurement Data in Roller Ski Skating
Md Zia Uddin
1
, Trine M. Seeberg
1
, Jan Kocbach
2
, Anders E. Liverud
1
, Victor Gonzalez
1
, Øyvind Sandbakk
2
and Frédéric Meyer
3,
*
Citation: Uddin, M.Z.; Seeberg, T.M.;
Kocbach, J.; Liverud, A.E.;
Gonzalez, V.; Sandbakk, Ø.; Meyer, F.
Estimation of Mechanical Power
Output Employing Deep Learning on
Inertial Measurement Data in Roller
Ski Skating. Sensors 2021, 21, 6500.
https://doi.org/10.3390/s21196500
Academic Editors: YangQuan Chen,
Subhas Mukhopadhyay,
Nunzio Cennamo, M. Jamal Deen,
Junseop Lee and Simone Morais
Received: 26 August 2021
Accepted: 24 September 2021
Published: 29 September 2021
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in
published maps and institutional affil-
iations.
Copyright: © 2021 by the authors.
Licensee MDPI, Basel, Switzerland.
This article is an open access article
distributed under the terms and
conditions of the Creative Commons
Attribution (CC BY) license (https://
creativecommons.org/licenses/by/
4.0/).
1
SINTEF Digital, 0373 Oslo, Norway; zia.uddin@sintef.no (M.Z.U.); Trine.Seeberg@sintef.no (T.M.S.);
anders.liverud@sintef.no (A.E.L.); victor.gonzalez@sintef.no (V.G.)
2
Centre for Elite Sports Research, Department of Neuromedicine and Movement Science,
Norwegian University of Science and Technology, 7491 Trondheim, Norway; jan.kocbach@gmail.com (J.K.);
oyvind.sandbakk@ntnu.no (Ø.S.)
3
Department of Informatics, University of Oslo, 0316 Oslo, Norway
* Correspondence: fredem@uio.no
Abstract:
The ability to optimize power generation in sports is imperative, both for understanding
and balancing training load correctly, and for optimizing competition performance. In this paper, we
aim to estimate mechanical power output by employing a time-sequential information-based deep
Long Short-Term Memory (LSTM) neural network from multiple inertial measurement units (IMUs).
Thirteen athletes conducted roller ski skating trials on a treadmill with varying incline and speed.
The acceleration and gyroscope data collected with the IMUs were run through statistical feature
processing, before being used by the deep learning model to estimate power output. The model was
thereafter used for prediction of power from test data using two approaches. First, a user-dependent
case was explored, reaching a power estimation within 3.5% error. Second, a user-independent case
was developed, reaching an error of 11.6% for the power estimation. Finally, the LSTM model was
compared to two other machine learning models and was found to be superior. In conclusion, the
user-dependent model allows for precise estimation of roller skiing power output after training the
model on data from each athlete. The user-independent model provides less accurate estimation;
however, the accuracy may be sufficient for providing valuable information for recreational skiers.
Keywords: cross country skiing; IMU; wearable sensors; LSTM; neural network
1. Introduction
Cross-country (XC) and roller skiing are endurance sports performed in varying terrain
with subsequent variations in speed, as well as both external and metabolic power [
1
–
3
].
The varying terrain of the courses used during training and competition induces periods
of very high intensity during uphill stints, and the ability to recover in downhill stints [
4
].
In addition, terrain, track and weather conditions influence the opposing forces and
constraints for producing power through poles and skis, which have a high impact on
skiing speed at a given metabolic intensity [
5
]. It is, therefore, not feasible to compare the
performance of athletes from day-to-day or from track-to-track by using speed or segment
time, as in many other sports, such as running or cycling. This highlights a need for metrics,
which can be used to track and compare performance, independently of track, terrain and
weather conditions.
Power meters, measuring the power output defined as the product of force and
velocity, are used extensively in cycling to quantitatively track changes in fitness and
performance [
6
]. However, while mechanical power can be directly measured on the bike
with force sensors (on the pedals, in the pedal or in the wheel), measurement of power
output in XC skiing is more complex, as force magnitude and direction must be measured
for both poles and skis [
7
]. Therefore, power output is commonly estimated using a power
Sensors 2021, 21, 6500. https://doi.org/10.3390/s21196500 https://www.mdpi.com/journal/sensors
