Sensor-Based Gym Physical Exercise Recognition: Data Acquisition and Experiments

Area of Science:

• Sports Science
• Biomedical Engineering
• Machine Learning

Background:

• Automated exercise tracking enhances motivation and health outcomes.
• Aerobic exercise trackers are common, but free weight exercise tracking remains manual.
• Accurate monitoring of weight training is crucial for balanced fitness programs.

Purpose of the Study:

• To develop a novel method for recognizing various gym-based free weight exercises.
• To utilize data from a single chest-mounted tri-axial accelerometer for exercise recognition.
• To confirm the feasibility of an automated system for comprehensive gym exercise analysis.

Main Methods:

• Data acquisition using a single chest-mounted tri-axial accelerometer.
• Development and testing of Long Short-Term Memory (LSTM) neural network models for exercise recognition.
• Experimentation with both single-muscle-group models and a universal model for all exercises.

Main Results:

• Demonstrated the feasibility of the proposed LSTM-based approach for recognizing a wide range of free weight exercises.
• Achieved promising results in distinguishing between different exercises using accelerometer data.
• Validated the effectiveness of both specialized and universal exercise recognition models.

Conclusions:

• The developed system offers a feasible solution for automated tracking and quantification of free weight exercises.
• This technology can contribute to comprehensive monitoring and analysis of gym-based workouts.
• Automated exercise recognition can enhance user experience by eliminating manual record-keeping.