Using Artificial Intelligence to Achieve Auxiliary Training of Table Tennis Based on Inertial Perception Data

Area of Science:

• Sports Science
• Biomechanical Analysis
• Machine Learning in Sports

Background:

• Wearable inertial sensors offer advantages over optical sensors for sports analysis, including cost, size, privacy, and mobility.
• Accurate recognition and evaluation of table tennis actions are crucial for effective athlete training and performance improvement.

Purpose of the Study:

• To develop a method for recognizing and evaluating human table tennis actions using wearable inertial sensors.
• To enhance auxiliary training for table tennis enthusiasts by providing fine-grained action analysis.

Main Methods:

• Collected human table tennis action data from professional and non-professional players using wearable inertial sensors.
• Extracted features from the sensor data.
• Proposed a multi-dimensional feature fusion convolutional neural network (CNN) for action recognition and evaluation.

Main Results:

• The proposed multi-dimensional feature fusion CNN achieved higher average recognition rates (0.17 and 0.16) compared to standard CNN and Inception-CNN on a non-professional test set.
• Demonstrated improved ability to distinguish between different table tennis actions and robust generalization performance.

Conclusions:

• The developed multi-dimensional feature fusion CNN is effective for recognizing and evaluating table tennis actions.
• This method provides a robust and accurate tool for auxiliary training, aiding table tennis players in improving their techniques.