Automatic Swimming Activity Recognition and Lap Time Assessment Based on a Single IMU: A Deep Learning Approach

Area of Science:

• Sports Science
• Biomechanical Analysis
• Machine Learning in Athletics

Background:

• Swimming performance analysis traditionally relies on complex video systems.
• Individual swimming technique and speed variations present significant analytical challenges.
• Objective, real-time biomechanical feedback is crucial for athlete development.

Purpose of the Study:

• To develop and validate a deep learning model for analyzing swimming performance using a single Inertial Measurement Unit (IMU).
• To accurately classify eight distinct swimming events (four strokes, rest, wall push, underwater, turns) across multiple velocities.
• To enable precise lap time computation for in-situ training monitoring.

Main Methods:

• Collected gyroscope and accelerometer data from 35 swimmers of varying skill levels.
• Developed a deep learning model accounting for inter- and intra-swimmer variability.
• Classified swimming events and velocities, calculating lap times with high temporal precision.

Main Results:

• Achieved an overall F1-score of 0.96 for classification accuracy.
• Demonstrated high temporal precision (0.02 s) for event detection.
• Validated lap time computation against video, yielding a mean absolute percentage error (MAPE) of 1.15% (start), 1% (middle), and 4.07% (end).

Conclusions:

• The deep learning model effectively analyzes swimming performance from a single IMU sensor.
• The system demonstrates potential as a training assistant for swimmers of all levels.
• High accuracy and temporal precision enable real-time, in-situ performance monitoring and feedback.