TL;DR
This paper introduces a "smart ski pole" equipped with inertial measurement units (IMUs) and load cells to capture detailed skiing technique data. This system achieves a remarkable 99.5% accuracy in recognizing five distinct cross-country skiing techniques, offering a valuable tool for athletes and coaches to analyze and improve performance.
What Problem Does This Paper Address?
Cross-country skiing is a complex sport where technique, propulsion, and overall performance are heavily influenced by the use of ski poles. While coaches and athletes recognize the importance of pole kinematics for improvement, obtaining comprehensive and accurate data has been challenging. Existing solutions often involve cumbersome wearable sensors or specialized equipment that can hinder natural movement or be prohibitively expensive. This paper aims to bridge this gap by developing a user-friendly, integrated smart ski pole that can easily collect crucial data like pole force, angle, and inertia, facilitating detailed analysis and technique refinement.
Methods
The core of this research is the development of a smart ski pole that integrates two key sensor types:
- Uniaxial Load Cell: This sensor measures the axial force applied to the ski pole, providing critical data on the propulsive force generated by the skier.
- Inertial Measurement Unit (IMU): Specifically, the ICM20602 IMU is used to capture acceleration and angular velocity data. This allows for the tracking of the pole's orientation, angle, and movement dynamics.
These sensors are housed within a specially designed grip and pole structure to minimize interference with natural skiing motion and ensure data integrity. The collected data is transmitted wirelessly (using LORA) to a computer for real-time analysis and storage.
For data analysis, the researchers employed:
- Sample t-test: Used to determine significant differences in skiing parameters (pole force, angle, time) between different skiers.
- Spearman Correlation Analysis: Applied to understand the relationships between various skiing metrics (e.g., pole force and speed, pole angle and speed) in skilled skiers.
- Support Vector Machine (SVM) based Action Recognition: This machine learning algorithm was trained using the extracted features from the sensor data (including time-domain and frequency-domain features) to classify five different cross-country skiing techniques.
Datasets
The study utilized data collected from three adult male participants who were experienced cross-country skiers. They performed the Double Poling (DP) technique on a flat 1000-m course. The data collected from these participants served as the basis for both the correlation analysis and the training/evaluation of the action recognition model. The paper does not mention any pre-existing public datasets for training or evaluation; the data appears to be newly collected for this study.
Key Insights and Ideas
- Ski Pole as a Rich Data Source: The research strongly validates the idea that ski poles are not just passive tools but active contributors to skiing performance, offering a wealth of kinematic and force data that can be leveraged for analysis.
- Integrated Sensing for Holistic Analysis: Combining load cells and IMUs in a single, unobtrusive device provides a more comprehensive understanding of skiing technique than either sensor alone.
- Correlation Between Force, Angle, and Speed: The study highlights significant correlations between pole force, pole angle, and skiing speed, suggesting that optimizing these parameters is crucial for maximizing performance.
- High Accuracy in Ski Technique Recognition: The development of an SVM-based model that achieves nearly perfect accuracy (99.5%) in identifying five distinct skiing techniques demonstrates the potential of this smart ski pole for automated performance analysis and feedback.
- User-Friendly Design: The emphasis on wireless transmission and a design that minimizes impact on the skier's natural movement makes this system practical for real-world training scenarios.
Key Results
- Significant Differences Between Skiers: A t-test revealed significant differences in pole force, pole angle, and pole time among the three participants, underscoring individual variations in technique.
- Strong Correlations: Spearman correlation analysis showed significant moderate correlations:
- Pole force and speed: r = 0.71
- Pole support angle and speed: r = 0.76
- Acceleration and axial load force: r = 0.55
- High Action Recognition Accuracy: The SVM model achieved an impressive 99.5% accuracy in recognizing five cross-country skiing techniques (DS, DP, KDP, G2, G5). This represents a significant improvement over previous similar systems.
Limitations and Future Work
The authors acknowledge several limitations and suggest avenues for future research:
- Sample Size: The study was conducted with only three participants. A larger and more diverse sample size is needed to generalize the findings.
- Simulated Environment: The experiments were performed on a flat course, simulating only a portion of the cross-country skiing process. Future work should incorporate more varied terrain and skiing stages.
- Future Analytical Models: Developing more complex analytical models that can encompass different skiing stages and a wider range of techniques is recommended.
- Broader Applications: The researchers envision future studies exploring more portable, practical, and affordable solutions for wider adoption across all levels of sports training. They also suggest the potential for integrating multi-sensor and wireless transmission modules for even more advanced portable performance measurement tools.
Summary updated