An effective swimming stroke recognition system utilizing deep learning based on inertial measurement units
Authors:
- 29 December 2022
- journal article
- research article
- Published by Taylor & Francis in Advanced Robotics
- Vol. 37 (7) , 467-479
Abstract
Machine learning (ML) and deep learning (DL) techniques have been applied to activity recognition in sports such as swimming with the help of inertial measurement units (IMU). In this study, we propose a DL-based swimming stroke recognition (SSR) system that can be used for swimming power-assisted exoskeletons by applying IMUs. The data of four swimming strokes are collected by 10 IMUs placed on the back, lumbar, upper limbs and lower limbs of swimmers. A Deep Convolutional Neural and Bidirectional Long Short-Term Memory network (DCNN-BiLSTM) is proposed for SSR and achieves a balanced accuracy of 96.27%. Then, the SSR system is optimized by selecting the quantity and layout of IMUs on the basis of an acceptable SSR accuracy with DCNN-BiLSTM in order to reduce the influence of IMUs on swimming. The optimized SSR system still has a balanced SSR accuracy of 93.44% while only applying two IMUs. Finally, two extra swimmers are invited to validate the optimized SSR system and receive a balanced accuracy of 92.67% and 92.94%. Although the results of this study, we can conclude that the proposed DCNN-BiLSTM can effectively recognize four target swimming strokes from different swimmers, which means that our system can be applied to exoskeletons. GRAPHICAL ABSTRACTKeywords
This publication has 48 references indexed in Scilit:
- Human activity recognition with smartphone sensors using deep learning neural networksExpert Systems with Applications, 2016
- A Deep Convolutional Neural Network for segmenting and classifying epithelial and stromal regions in histopathological imagesNeurocomputing, 2016
- Deep Convolutional and LSTM Recurrent Neural Networks for Multimodal Wearable Activity RecognitionSensors, 2016
- Inertial Sensor Technology for Elite Swimming Performance Analysis: A Systematic ReviewSensors, 2015
- Classification of kinematic swimming data with emphasis on resource consumption2013
- Feature Extraction and Reduction of Wavelet Transform Coefficients for EMG Pattern ClassificationElektronika ir Elektrotechnika, 2012
- Swimming performance and technique evaluation with wearable acceleration sensorsPervasive and Mobile Computing, 2012
- Driver Drowsiness Classification Using Fuzzy Wavelet-Packet-Based Feature-Extraction AlgorithmIEEE Transactions on Biomedical Engineering, 2010
- ISB recommendation on definitions of joint coordinate systems of various joints for the reporting of human joint motion—Part II: shoulder, elbow, wrist and handJournal of Biomechanics, 2005
- ISB recommendation on definitions of joint coordinate system of various joints for the reporting of human joint motion—part I: ankle, hip, and spineJournal of Biomechanics, 2002