In the current market, successful fitness tracking devices utilize heart rate and GPS to determine performance. These devices are useful, but don't extensively classify stationary exercise. This paper proposes a modern approach for tuning and investigating optimal neural network types on stationary exercises using Inertial Measurement Units (IMUs). Using three IMUs located on the ankle, waist, and wrist, data is collected to map the body as it moves during the stationary physical activity. A novel five-stage deep learning tuning system was written and deployed to classify user movement as one of three classes: air squats, jumping jacks, and kettlebell swings. It was determined that the ConvLSTM2D type is the most accurate and second fastest for training stationary exercise classification. Tracking of human movement extends to realms outside of fitness such as robotics, medical and military applications.
Heroy, Andrew M.; Gill, Zackary; Sprague, Samantha; Stroud, David; and Santerre, John
"Stationary Exercise Classification using IMUs and Deep Learning,"
SMU Data Science Review: Vol. 3:
1, Article 1.
Available at: https://scholar.smu.edu/datasciencereview/vol3/iss1/1
Creative Commons License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License
Dynamics and Dynamical Systems Commons, Exercise Science Commons, Physical Therapy Commons, Sports Sciences Commons