An Artificial Intelligence Exercise Coaching Mobile App: Development and Randomized Controlled Trial to Verify Its...

Area of Science:

• Biomedical Engineering
• Computer Science
• Sports Science

Background:

• Physical inactivity increases disease risk; incorrect exercise form can lead to injury.
• Existing remote coaching and posture correction solutions are often expensive or inefficient.
• Mobile devices offer a potential platform for accessible, AI-powered fitness guidance.

Purpose of the Study:

• To develop a mobile application using deep neural networks for personalized squat posture feedback.
• To leverage deep learning to mimic expert assessment of exercise form.
• To evaluate the app's effectiveness in improving squat technique compared to standard exercise videos.

Main Methods:

• Trained a deep learning model on over 20,000 expert-annotated squat videos using pose estimation and video classification.
• Developed the 'Home Alone Exercise' mobile app for real-time posture analysis.
• Conducted a 2-week randomized controlled trial comparing app users (EXP) with a control group (CTL) using only exercise videos.

Main Results:

• The EXP group showed significant improvements in squat posture scores (P=.001) and knee joint angles (P=.02-.03) after 2 weeks.
• The CTL group exhibited no significant changes in squat posture or knee joint angles.
• The app effectively identified and guided users toward correct squat execution.

Conclusions:

• The mobile workout assistant provides cost-effective, self-guided feedback for squat exercises.
• Users trained with the app demonstrated faster learning and better understanding of nuanced exercise details.
• This AI-powered approach offers an efficient alternative to in-person training for improving exercise form.