Table Tennis Tutor: Forehand Strokes Classification Based on Multimodal Data and Neural Networks.

Area of Science:

• Sports Science
• Human-Computer Interaction
• Machine Learning

Background:

• Beginner table tennis players need continuous feedback for technique development.
• Limited access to coaches and expensive equipment hinder real-time training.
• Sensor technology and machine learning offer solutions for automated skill assessment.

Purpose of the Study:

• Introduce the Table Tennis Tutor (T3), a multi-sensor system for real-time feedback in table tennis.
• Detect forehand stroke errors in beginners using sensor data.
• Evaluate the effectiveness of smartphone sensors and Microsoft Kinect for stroke analysis.

Main Methods:

• Developed the Table Tennis Tutor (T3) system using smartphone sensors and a Microsoft Kinect.
• Collected a dataset of correct and incorrect forehand strokes.
• Trained a recurrent neural network for classifying stroke accuracy.
• Validated sensor combinations: smartphone only, Kinect only, and combined.

Main Results:

• The combined system of smartphone sensors and Kinect demonstrated improved precision in detecting forehand stroke errors.
• Smartphone sensors alone showed lower performance compared to the Kinect.
• Expert interviews indicated positive perceptions of T3 as a training aid.

Conclusions:

• The T3 system, leveraging multimodal sensor data, shows promise for providing effective real-time feedback to table tennis beginners.
• Combining smartphone sensors with Kinect enhances the accuracy of automated technique analysis.
• Future implementations can benefit from expert coach insights for improved training system development.