An approach to sport activities recognition based on an inertial sensor and deep learning

• DOI: 10.1016/j.sna.2022.113773
• Publication Date: 2022-07-22T03:25:18Z
• Authors: Grzegorz Pajak; Pascal Krutz; Justyna Patalas-M...; Matthias Rehm; Iwona Pajak; Martin Dix
• Keywords: Activity Recognition; Training set; Kernel (algebra); Data set

In recent years, due to changing the human lifestyle, number of sport trainers has been increased. The conventional classifiers as Naive Bayes (NB), Decision Trees (DT) and Convolutional Neural Networks (CNNs) can be used in this domain recognize count sports activities subjects provide them qualified feedback. This paper uses literature studies selected activities, namely squats, pull-ups dips dataset based on three UWB sensors with additional inertial data, which contains reduced data set consisting 17 training sets next for CNN 1444 samples describing exercises 2024 breaks, were grouped ratio 70:15:15. recognition accuracy NB DT 89.4 92.9 accordingly. Next, extensive performance analysis experiments different kernel sizes, filters single dual layer networks was carried out. Moreover, innovative model form combination several forming Ensemble Network (ENN) created. at level 94.81 exceeded 95% ENN. proposed prototype measurement system acquisition platform highlighted great potential privacy-training system.