Automatic Exercise Recognition with Machine Learning Chapter uri icon

abstract

  • Springer Nature Switzerland AG 2020. Although most individuals understand the importance of regular physical activity, many still lead mostly sedentary lives. The use of smartphones and fitness trackers has mitigated this trend some, as individuals are able to track their physical activity; however, these devices are still unable to reliably recognize many common exercises. To that end, we propose a system designed to recognize sit ups, bench presses, bicep curls, squats, and shoulder presses using accelerometer data from a smartwatch. Additionally, we evaluate the effectiveness of this recognition in a real-time setting by developing and testing a smartphone application built on top of this system. Our system recognized these activities with overall F-measures of 0.94 and 0.87 in a controlled environment and real-time setting respectively. Both users who were and who were not regularly physically active responded positively to our system, noting that our system would encourage them to continue or start exercising regularly.

author list (cited authors)

  • Mendiola, V., Doss, A., Adams, W., Ramos, J., Bruns, M., Cherian, J., ... Hammond, T.

citation count

  • 2

complete list of authors

  • Mendiola, Victor||Doss, Abnob||Adams, Will||Ramos, Jose||Bruns, Matthew||Cherian, Josh||Kohli, Puneet||Goldberg, Daniel||Hammond, Tracy

Book Title

  • Precision Health and Medicine

publication date

  • January 2020