Abstract
This preliminary study examined time domain, frequency domain, and time-frequency domain features for the appropriate classification of four different exercises. Thirty bench presses, squats, crunches, and walking trials were conducted, and data from two other sensor locations were considered (accelerometer data on the right wrist and left ankle). Time, frequency, and time-frequency features were examined across trials with differences in processing methods applied. Multiple classification algorithms were used and validated using 5-fold self-validation and validation against new user data. Results indicated that the features and algorithms investigated could accurately classify exercise routines. Additional exercises and larger data sets are being developed for a more robust study.