Wearable Fall Detection Based on Motion Signals Using Hybrid Deep Residual Neural Network
Issued Date
2022-01-01
Resource Type
ISSN
03029743
eISSN
16113349
Scopus ID
2-s2.0-85142697973
Journal Title
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume
13651 LNAI
Start Page
216
End Page
224
Rights Holder(s)
SCOPUS
Bibliographic Citation
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) Vol.13651 LNAI (2022) , 216-224
Suggested Citation
Mekruksavanich S., Jantawong P., Hnoohom N., Jitpattanakul A. Wearable Fall Detection Based on Motion Signals Using Hybrid Deep Residual Neural Network. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) Vol.13651 LNAI (2022) , 216-224. 224. doi:10.1007/978-3-031-20992-5_19 Retrieved from: https://repository.li.mahidol.ac.th/handle/20.500.14594/85118
Title
Wearable Fall Detection Based on Motion Signals Using Hybrid Deep Residual Neural Network
Author's Affiliation
Other Contributor(s)
Abstract
There have been several approaches for wearable fall detection devices during the last twenty years. The majority of technologies relied on machine learning. Although the given findings appear that the issue is practically addressed, critical problems remain about feature extraction and selection. In this research, the constraint of machine learning on feature extraction is addressed by including a hybrid convolutional operation in our proposed deep residual network, called the DeepFall model. The proposed network automatically generates high-level motion signal characteristics that can be utilized to track falls and everyday activities. FallAllD dataset, a publicly available standard dataset for fall detection that gathered motion signals of falls and other events, was utilized to analyze the proposed network. We performed investigations using a 5-fold cross-validation technique to determine overall accuracy and F-measure. The experimental outcomes show that the proposed DeepFall performs better accuracy (95.19%) and F-measure (92.79%) than the state-of-the-art baseline deep learning networks.