Convolutional long short-term memory model for recognizing postures from wearable sensor

Research output: Contribution to journalConference article

Abstract

This research investigates the feasibility and viability of applying Deep Neural Networks (DNN) to improve performance with respect to posture recognition based on multi-channel motion data from Wearable Sensors (WS). The authors use the recognition of posture that can be linked to risk of Musculoskeletal Disorder (MSD)- among construction workers as the testbed. The proposed approach is based on the use of a DNN model integrating Convolutional Neural Network (CNN) and Long short-term memory (LSTM) that can achieve automated feature engineering and sequential pattern detection. The model performance was evaluated using datasets collected from four construction workers. The proposed model outperformed baseline CNN and LSTM models. Under the personalized modelling approach, it improved recognition performance by 3% from the benchmark Machine Learning models; the improvement is 2% for generalized modelling approach. The proposed model achieves high-performance posture recognition, which facilitates the MSD prevention in construction through monitoring injury-related postures.

Original languageEnglish (US)
JournalCEUR Workshop Proceedings
Volume2394
StatePublished - Jan 1 2019
Event26th International Workshop on Intelligent Computing in Engineering, EG-ICE 2019 - Leuven, Belgium
Duration: Jun 30 2019Jul 3 2019

Fingerprint

Neural networks
Wearable sensors
Long short-term memory
Testbeds
Learning systems
Monitoring
Deep neural networks

All Science Journal Classification (ASJC) codes

  • Computer Science(all)

Cite this

@article{9e7a6fd841d349caa4dc543b28a79ed5,
title = "Convolutional long short-term memory model for recognizing postures from wearable sensor",
abstract = "This research investigates the feasibility and viability of applying Deep Neural Networks (DNN) to improve performance with respect to posture recognition based on multi-channel motion data from Wearable Sensors (WS). The authors use the recognition of posture that can be linked to risk of Musculoskeletal Disorder (MSD)- among construction workers as the testbed. The proposed approach is based on the use of a DNN model integrating Convolutional Neural Network (CNN) and Long short-term memory (LSTM) that can achieve automated feature engineering and sequential pattern detection. The model performance was evaluated using datasets collected from four construction workers. The proposed model outperformed baseline CNN and LSTM models. Under the personalized modelling approach, it improved recognition performance by 3{\%} from the benchmark Machine Learning models; the improvement is 2{\%} for generalized modelling approach. The proposed model achieves high-performance posture recognition, which facilitates the MSD prevention in construction through monitoring injury-related postures.",
author = "Junqi Zhao and Obonyo, {Esther Adhiambo}",
year = "2019",
month = "1",
day = "1",
language = "English (US)",
volume = "2394",
journal = "CEUR Workshop Proceedings",
issn = "1613-0073",
publisher = "CEUR-WS",

}

Convolutional long short-term memory model for recognizing postures from wearable sensor. / Zhao, Junqi; Obonyo, Esther Adhiambo.

In: CEUR Workshop Proceedings, Vol. 2394, 01.01.2019.

Research output: Contribution to journalConference article

TY - JOUR

T1 - Convolutional long short-term memory model for recognizing postures from wearable sensor

AU - Zhao, Junqi

AU - Obonyo, Esther Adhiambo

PY - 2019/1/1

Y1 - 2019/1/1

N2 - This research investigates the feasibility and viability of applying Deep Neural Networks (DNN) to improve performance with respect to posture recognition based on multi-channel motion data from Wearable Sensors (WS). The authors use the recognition of posture that can be linked to risk of Musculoskeletal Disorder (MSD)- among construction workers as the testbed. The proposed approach is based on the use of a DNN model integrating Convolutional Neural Network (CNN) and Long short-term memory (LSTM) that can achieve automated feature engineering and sequential pattern detection. The model performance was evaluated using datasets collected from four construction workers. The proposed model outperformed baseline CNN and LSTM models. Under the personalized modelling approach, it improved recognition performance by 3% from the benchmark Machine Learning models; the improvement is 2% for generalized modelling approach. The proposed model achieves high-performance posture recognition, which facilitates the MSD prevention in construction through monitoring injury-related postures.

AB - This research investigates the feasibility and viability of applying Deep Neural Networks (DNN) to improve performance with respect to posture recognition based on multi-channel motion data from Wearable Sensors (WS). The authors use the recognition of posture that can be linked to risk of Musculoskeletal Disorder (MSD)- among construction workers as the testbed. The proposed approach is based on the use of a DNN model integrating Convolutional Neural Network (CNN) and Long short-term memory (LSTM) that can achieve automated feature engineering and sequential pattern detection. The model performance was evaluated using datasets collected from four construction workers. The proposed model outperformed baseline CNN and LSTM models. Under the personalized modelling approach, it improved recognition performance by 3% from the benchmark Machine Learning models; the improvement is 2% for generalized modelling approach. The proposed model achieves high-performance posture recognition, which facilitates the MSD prevention in construction through monitoring injury-related postures.

UR - http://www.scopus.com/inward/record.url?scp=85069161664&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85069161664&partnerID=8YFLogxK

M3 - Conference article

AN - SCOPUS:85069161664

VL - 2394

JO - CEUR Workshop Proceedings

JF - CEUR Workshop Proceedings

SN - 1613-0073

ER -