Human Activity Recognition (HAR) can be performed through an actigraph that: 1) records 3D accelerometric and gyroscopic signals, 2) classifies different activities of daily living, such as walking (ground level, uphill, downhill), ascending or descending stairs, upright quiet standing, and sitting. The aim of this work is to introduce a deep learning classification approach, based on Long Short-Term Memory (LSTM) neural networks, that can be easily implemented on a HAR actigraph. Among 360 different models, the best LSTM was chosen and its properties described. The classifier accuracy ranges from 92% (for uphill walking) to 100% (for sitting). Differently from previously described HAR classifiers, the proposed classifier requires neither signal pre-processing, nor feature extraction and selection.
Human Activity Recognition through Wearable Sensors: a Deep Learning Approach / Fortunato, D.; Ghislieri, M.; Rosati, S.; Balestra, G.; Knaflitz, M.; Agostini, V.. - ELETTRONICO. - (2020), pp. 368-371. (Intervento presentato al convegno 7th National Congress of Bioengineering, GNB 2020 tenutosi a Trieste, Italy nel 9-11 June 2020).
Human Activity Recognition through Wearable Sensors: a Deep Learning Approach
Fortunato D.;Ghislieri M.;Rosati S.;Balestra G.;Knaflitz M.;Agostini V.
2020
Abstract
Human Activity Recognition (HAR) can be performed through an actigraph that: 1) records 3D accelerometric and gyroscopic signals, 2) classifies different activities of daily living, such as walking (ground level, uphill, downhill), ascending or descending stairs, upright quiet standing, and sitting. The aim of this work is to introduce a deep learning classification approach, based on Long Short-Term Memory (LSTM) neural networks, that can be easily implemented on a HAR actigraph. Among 360 different models, the best LSTM was chosen and its properties described. The classifier accuracy ranges from 92% (for uphill walking) to 100% (for sitting). Differently from previously described HAR classifiers, the proposed classifier requires neither signal pre-processing, nor feature extraction and selection.File | Dimensione | Formato | |
---|---|---|---|
GNB2020_HARDeepLearning_Revised.pdf
accesso riservato
Descrizione: Full-text
Tipologia:
2. Post-print / Author's Accepted Manuscript
Licenza:
Non Pubblico - Accesso privato/ristretto
Dimensione
572.5 kB
Formato
Adobe PDF
|
572.5 kB | Adobe PDF | Visualizza/Apri Richiedi una copia |
Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/11583/2979251