TY - JOUR
T1 - Quantifying swimming activities using accelerometer signal processing and machine learning
T2 - A pilot study
AU - Qin, Xiong
AU - Song, Yadong
AU - Zhang, Guanqun
AU - Guo, Fan
AU - Zhu, Weimo
N1 - Funding Information:
This project was partially supported by a gift fund from Lifesense/Transtek Health Inc. to the University of Illinois at Urbana-Champaign, USA.
Publisher Copyright:
© 2021 Elsevier Ltd
PY - 2022/1
Y1 - 2022/1
N2 - Aerobic exercises on land could be quantified and tracked objectively, but swimming style recognition has remained unexplored. Taking the advantages of signal processing and machine learning on acceleration signals, the purpose of this study was, by analyzing swimming accelerometer data, to explore a set of algorithm in tracking swimming activities, including recognizing swimming styles, counting time and counting strokes in each style. A total of 17 participants (9 females) from the swimming team of the Southeast University of China was recruited. They performed breaststroke, front crawl, backstroke and butterfly, four 50-meter-lap each, with an ActiGraph GT9X inertia measurement unit on wrist of their preferred side. Overall, 78.7 ± 14.6, 148.5 ± 21.7, 151.2 ± 14.4, 98 ± 16.3 strokes were performed and evaluated on breaststroke, front crawl, backstroke and butterfly, respectively. In classification, three classifiers were examined and the result showed that support vector machine (SVM) provided the best accuracy of classification (over 99%). In time counting, the accuracy was over 99% and in stroke counting, the overall single-lap accuracy rate was 93.3%. In conclusion, with a combination of an objective measure and machine-learning algorithm, tracking swimming activities, including swimming style classification, counting swimming time and strokes, by a accelerometer becomes possible.
AB - Aerobic exercises on land could be quantified and tracked objectively, but swimming style recognition has remained unexplored. Taking the advantages of signal processing and machine learning on acceleration signals, the purpose of this study was, by analyzing swimming accelerometer data, to explore a set of algorithm in tracking swimming activities, including recognizing swimming styles, counting time and counting strokes in each style. A total of 17 participants (9 females) from the swimming team of the Southeast University of China was recruited. They performed breaststroke, front crawl, backstroke and butterfly, four 50-meter-lap each, with an ActiGraph GT9X inertia measurement unit on wrist of their preferred side. Overall, 78.7 ± 14.6, 148.5 ± 21.7, 151.2 ± 14.4, 98 ± 16.3 strokes were performed and evaluated on breaststroke, front crawl, backstroke and butterfly, respectively. In classification, three classifiers were examined and the result showed that support vector machine (SVM) provided the best accuracy of classification (over 99%). In time counting, the accuracy was over 99% and in stroke counting, the overall single-lap accuracy rate was 93.3%. In conclusion, with a combination of an objective measure and machine-learning algorithm, tracking swimming activities, including swimming style classification, counting swimming time and strokes, by a accelerometer becomes possible.
KW - Accelerometer
KW - Classification
KW - Stroke count
KW - Tracking
UR - http://www.scopus.com/inward/record.url?scp=85115026706&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85115026706&partnerID=8YFLogxK
U2 - 10.1016/j.bspc.2021.103136
DO - 10.1016/j.bspc.2021.103136
M3 - Article
AN - SCOPUS:85115026706
SN - 1746-8094
VL - 71
JO - Biomedical Signal Processing and Control
JF - Biomedical Signal Processing and Control
M1 - 103136
ER -