Wearable Fall Detection Based on Motion Signals Using Hybrid Deep Residual Neural Network
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. I...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Conference or Workshop Item |
| Published: |
2023
|
| Subjects: | |
| Online Access: | https://repository.li.mahidol.ac.th/handle/123456789/85118 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Mahidol University |
| Summary: | 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. |
|---|