A multi-sensor system for automatic analysis of classical cross-country skiing techniques

Seeberg, Trine M.; Tjønnås, Johannes; Rindal, Ole Marius Hoel; Haugnes, Pål; Dalgard, Steffen Harald; Sandbakk, Øyvind

doi:10.1007/s12283-017-0252-z

Cited by 27 publications

(40 citation statements)

References 16 publications

(31 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The sensor data were recorded using a previously described multi-sensor system [ 12 ] that can synchronize data from seven IMUs with an accuracy of at least 30 ms. However, in this study only data from two of the IMUs were used: the accelerometer data of the chest sensor and the gyroscope data from one arm.…”

Section: Methodsmentioning

confidence: 99%

“…The IMU-sensor on the arm has the coordinate system as indicated in Figure 1 . To compensate for any misalignments in the mounting of the arm sensor, the coordinate system is calibrated based on the assumption that the axis with the most angular motion is perpendicular to the sagittal plane [ 12 ]. The calibrated gyroscope data were then filtered with a Gaussian low-pass filter, with 0.25 s (5 samples) standard deviation in the time domain.…”

Section: Methodsmentioning

confidence: 99%

“…We have earlier classified classical sub-techniques on snow using hard decision rules on data from IMUs on arms and legs [ 12 ], resulting in classification sensitivity and precision of 99% to 100% for DIA, DK and DP. However, only three main classes were classified with the rest of the data classified as “unknown”.…”

Section: Introductionmentioning

confidence: 99%

“…The choice of using only two sensors is to reduce the sensitivity of data loss and to make it more convenient to wear, for example, by integrating one IMU with the heart rate sensor on the chest and the other in a sports watch on the arm. Compared to the previous methodologies [ 6 , 7 , 8 , 9 , 11 , 12 ], we have refined the division of classes by adding HRB, TCK and TRN as separate classes and introduced the classifications of transitions to DIA (tDIA) and from DIA (fDIA). Our hypothesis was that a machine learning algorithm with a robust cycle detection and a refinement of the sub-techniques could classify sub-techniques with significantly higher accuracy than in previously reported studies.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Automatic Classification of Sub-Techniques in Classical Cross-Country Skiing Using a Machine Learning Algorithm on Micro-Sensor Data

Rindal

Seeberg

Tjønnås

et al. 2017

Sensors

Self Cite

View full text Add to dashboard Cite

The automatic classification of sub-techniques in classical cross-country skiing provides unique possibilities for analyzing the biomechanical aspects of outdoor skiing. This is currently possible due to the miniaturization and flexibility of wearable inertial measurement units (IMUs) that allow researchers to bring the laboratory to the field. In this study, we aimed to optimize the accuracy of the automatic classification of classical cross-country skiing sub-techniques by using two IMUs attached to the skier’s arm and chest together with a machine learning algorithm. The novelty of our approach is the reliable detection of individual cycles using a gyroscope on the skier’s arm, while a neural network machine learning algorithm robustly classifies each cycle to a sub-technique using sensor data from an accelerometer on the chest. In this study, 24 datasets from 10 different participants were separated into the categories training-, validation- and test-data. Overall, we achieved a classification accuracy of 93.9% on the test-data. Furthermore, we illustrate how an accurate classification of sub-techniques can be combined with data from standard sports equipment including position, altitude, speed and heart rate measuring systems. Combining this information has the potential to provide novel insight into physiological and biomechanical aspects valuable to coaches, athletes and researchers.

show abstract

Section: Methodsmentioning

confidence: 99%

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Automatic Classification of Sub-Techniques in Classical Cross-Country Skiing Using a Machine Learning Algorithm on Micro-Sensor Data

Rindal

Seeberg

Tjønnås

et al. 2017

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…To better understand overall performance, as well as the physical and technical demands in XC sit-skiing, continuous measurements of speed, and the corresponding power output, kinematic patterns (i.e., cycle rate and length) and heart rate (HR), as a proxy of metabolic intensity, for the various terrains throughout training and competitions would provide important information. In this context, the utilization of combined Global Navigation Satellite Systems (GNSS) and micro-sensor technology in the field has shown potential in ablebodied XC skiers (Sandbakk et al, 2011(Sandbakk et al, , 2016Marsland et al, 2012Marsland et al, , 2015Marsland et al, , 2017Marsland et al, , 2018Seeberg et al, 2017;Solli et al, 2018) and could provide valuable insight also in XC sit-skiers.…”

Section: Introductionmentioning

confidence: 99%

Development of a Framework for the Investigation of Speed, Power, and Kinematic Patterns in Para Cross-Country Sit-Skiing: A Case Study of an LW12 Athlete

Baumgart

Haugnes

Bardal

et al. 2019

Front. Sports Act. Living

Self Cite

View full text Add to dashboard Cite

Objective: To develop a framework for the investigation of speed, power, and kinematic patterns across varying terrain in crosscountry (XC) sit-skiing, and to test this framework in a XC sit-skier of the LW12 class during high-(HIT) and low-intensity (LIT) endurance training. Methods: One XC sit-skiing athlete of the LW12 class with a single above-the-knee amputation was equipped with a GNSS enabled sports watch with integrated barometry and heart rate monitoring (peak heart rate: 195 beats•min −1), and an inertial measurement unit. After a warm-up, he performed two 20-m maximal speed tests on a flat and an uphill section to determine maximal speed and power, followed by skiing 5.75 km at both LIT and HIT in varying terrain. Results: 51, 28, and 21% of the time during HIT and 53, 28, and 19% of the time during LIT were spent in uphill, flat and downhill terrain, respectively. Maximal speed in the uphill and flat section was 4.0 and 6.2 m•s −1 , respectively, and the corresponding maximal power output 342 and 252 W. The % of maximal speed did not differ between the uphill and the flat section (HIT: 66 vs. 67%, LIT: 47 vs. 50%), whereas the % of maximal power output was lower in the uphill than flat section (HIT: 65 and 80%, LIT: 46 and 58%). Still, the absolute power output was slightly higher in the uphill than the flat section (HIT: 222 vs. 201 W, LIT: 156 vs. 145 W). Furthermore, cycle rate was significantly higher during HIT than LIT (60-61 vs. 45-55 cycles•min −1 , across all terrains, all p < 0.03), while cycle length was longer in the uphill terrain (3.0 vs. 2.6 m, p < 0.001). Furthermore, the % of peak heart rate was significantly higher in HIT than LIT (90 vs. 78, 85 vs. 67, and 88 vs. 66%, respectively, in the uphill, flat and downhill terrain, all p < 0.001). Conclusions: Here, we present a new integrative framework for future investigations of performance, technical and physical demands in XC sit-skiing. In this case study, the Baumgart et al. Framework for Investigating Para XC Sit-Skiing increase in speed from LIT to HIT was due to increases in cycle rate in all terrains, while cycle length was less affected. Although the absolute power output was slightly higher in the uphill compared to the flat section both for HIT and LIT, the athlete worked closer to his maximum power output in the flat section.

show abstract

Machine Learning‐Enhanced Triboelectric Sensing Application

Shang,

Chen,

Cai

et al. 2024

Adv Materials Technologies

View full text Add to dashboard Cite

Triboelectric nanogenerator (TENG) has become a promising candidate for wearable energy harvesting and self‐powered sensing systems. However, processing large amounts of data imposes a computing power barrier for practical application. Machine learning‐assisted self‐powered sensors based on TENG have been widely used in data‐driven applications due to their excellent characteristics such as no additional power supply, high sensing accuracy, low cost, and good biocompatibility. This work comprehensively reviews the latest progress in machine learning (ML)‐assisted TENG‐based sensors. The future challenges and opportunities are discussed. First, the fundamental principles including the working mode of ML‐assisted TENG‐based sensor and common algorithms are systematically and comprehensively illustrated, which emphasizes the algorithm definition and principle. Subsequently, the progress of ML methods in the field of TENG‐based sensors is further reviewed, summarizing the advantages and disadvantages of various algorithms in practical examples, and providing guidance and suggestions on how to choose the appropriate methods. Finally, the prospects and challenges of ML‐assisted TENG‐based sensors is summarized. Directions and important insights for the future development of TENG and AI integration is provided.

show abstract

A multi-sensor system for automatic analysis of classical cross-country skiing techniques

Cited by 27 publications

References 16 publications

Automatic Classification of Sub-Techniques in Classical Cross-Country Skiing Using a Machine Learning Algorithm on Micro-Sensor Data

Automatic Classification of Sub-Techniques in Classical Cross-Country Skiing Using a Machine Learning Algorithm on Micro-Sensor Data

Development of a Framework for the Investigation of Speed, Power, and Kinematic Patterns in Para Cross-Country Sit-Skiing: A Case Study of an LW12 Athlete

Machine Learning‐Enhanced Triboelectric Sensing Application

Contact Info

Product

Resources

About