TL;DR
This paper introduces an AIoT-based system using a wearable sensor on the ankle to track and analyze soccer motions for training and talent selection. The system offers a cost-effective, real-time alternative to traditional video analysis, achieving high accuracy in distinguishing between different actions and skill levels.
What Problem Does This Paper Address?
Traditional methods for analyzing athlete performance in sports like soccer often rely on subjective coaching and expensive, cumbersome video analysis setups. These methods struggle with real-time feedback, require significant computational resources, and can miss subtle, rapid movements crucial for skill development. Specifically for soccer, the complex and varied motions of the feet are challenging to capture and analyze objectively. This paper aims to bridge this gap by developing an accessible, IoT-enabled system that provides precise, real-time motion tracking and skill assessment for foot-related sports.
Methods
The core of the system is an Artificial Intelligence of Things (AIoT) paradigm, integrating Artificial Intelligence (AI) with the Internet of Things (IoT). The system comprises three main components:
- Customized Wireless Wearable Sensing Device (WWSD): A small, lightweight device worn on the ankle, equipped with a 3-axis gyroscope and accelerometer (specifically, a Bosch BMI160 IMU chip). This device captures raw motion data.
- Mobile Application: This app connects to the WWSD via Bluetooth Low Energy (BTLE), receives the motion data, visualizes it in real-time, and transmits it to a cloud-based platform.
- Cloud-Based Data Processing Platform: This platform receives data from the mobile app, preprocesses it (including a 3-point moving average filter for noise reduction and signal segmentation), and applies an ankle-based attitude angle analysis model.
A key methodological innovation is the development of an ankle-based attitude angle model. This model extracts more meaningful features by calculating Euler angles (yaw, pitch, roll) from the sensor data, which are crucial for understanding foot movements in soccer. Principal Component Analysis (PCA) is then used for dimensionality reduction to balance system complexity and accuracy. Finally, a Support Vector Machine (SVM) classification algorithm is employed to recognize different soccer motions (passing and shooting) and assess skill levels (elite vs. amateur).
Datasets
The study utilized a dataset collected from 11 male soccer players (5 elites and 6 amateurs) at the Chinese Football Association (CFA) Youth Academy and the Sports Center of South-Central University. Each participant performed 20 instances of passing and 20 instances of shooting. The paper does not mention any pre-existing public datasets being used for training or evaluation, suggesting the dataset was collected specifically for this research.
Key Insights and Ideas
- Ankle-centric sensing is crucial for foot-based sports: Unlike many existing wearable sensor applications that focus on the wrist or other body parts, this study highlights the importance of ankle-mounted sensors for accurately capturing the nuances of soccer movements.
- Attitude angles provide richer features: Simply using raw acceleration and angular velocity data is insufficient. Deriving attitude angles (yaw, pitch, roll) from the sensor data significantly enhances the system's ability to differentiate between complex motions and skill levels.
- AIoT offers a practical solution: The integration of AI and IoT provides a cost-effective, real-time, and accessible system that can democratize advanced performance analysis, moving beyond traditional, resource-intensive methods.
- Transition from experience-driven to data-driven sports analysis: The system represents a significant step towards making sports training and talent selection more objective and data-informed, reducing reliance on subjective expert opinions.
Key Results
The proposed AIoT system demonstrated strong performance:
- Motion Recognition: The system achieved an average accuracy of 87.1% for classifying different soccer motions (passing and shooting). Specifically, passing recognition was 85.7% and shooting recognition was 88.5%.
- Talent Selection: The system was highly effective in distinguishing between elite and amateur players. The accuracy for talent selection reached 93% for shooting actions, indicating a significant skill gap that the system can identify. The passing skill level assessment also showed good accuracy, with elites identified at 88.7% and amateurs at 85.1%.
- Comparison with other algorithms: The SVM-based classifier, particularly the one incorporating attitude angle features, outperformed KNN and Decision Tree algorithms in both motion recognition and talent selection tasks.
Limitations and Future Work
- Dataset Size and Diversity: The study used a relatively small sample size (11 participants) and focused on male players. Future work could expand the dataset to include more participants, different age groups, and female athletes to improve generalizability.
- Computational Load and Storage: While the system is less resource-intensive than video analysis, the paper acknowledges that the cloud system's storage performance and computing speed could be improved.
- Edge Computing Integration: To further enhance real-time response and reduce reliance on cloud infrastructure, the authors suggest integrating edge computing into the system. This would allow for more processing to occur closer to the data source (i.e., on the mobile device or a local gateway).
- Extension to Other Sports: The authors propose that the system's architecture and methodology can be easily extended to analyze motions and skill levels in other foot-related sports like taekwondo, tumbling, and gymnastics.
Summary updated