Recognition and Repetition Counting for Local Muscular Endurance Exercises in Exercise-Based Rehabilitation: A...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Cardiology

Background:

• Cardiac rehabilitation programs often involve prescribed exercises with specific repetition counts.
• Remote patient monitoring is crucial for independent exercise adherence and recovery.
• Accurate, automated exercise recognition and counting from wearable sensors are needed.

Purpose of the Study:

• To compare Convolutional Neural Networks (CNNs) against traditional methods for exercise recognition and repetition counting.
• To evaluate the effectiveness of AlexNet-based CNN models for local muscular endurance exercises.
• To introduce a novel, single CNN approach for simultaneous exercise recognition and repetition counting.

Main Methods:

• Implementation and comparison of supervised machine learning (ML) and peak detection algorithms.
• Development and testing of AlexNet-based Convolutional Neural Networks (CNNs) using wearable sensor data.
• Validation of CNN performance for both exercise recognition and repetition counting tasks.

Main Results:

• CNNs achieved superior performance compared to traditional supervised ML and peak detection methods.
• Overall F1-score for exercise recognition reached 97.18% using CNNs.
• CNNs demonstrated high accuracy in repetition counting, with ±1 error in 90% of observed sets.

Conclusions:

• Convolutional Neural Networks offer a highly effective solution for automated exercise recognition and repetition counting in cardiac rehabilitation.
• The proposed single CNN method for both tasks is novel and efficient.
• The public release of the INSIGHT-LME dataset will foster further research in this domain.