Fall detection - prediction based on deep learning techniques
Loading...
Date
2025-10-08
Authors
Αρβανιτόπουλος, Λέανδρος Νικόλαος
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
There are many falls that happen throughout the world, that mainly affect the elderly population. Aside from being a traumatic experience, falls are a serious threat, since a large percentage of falls are lethal. Smartphones are truly ubiquitous in our current age, even among older people, all the while their computational power keeps increasing. The rise of more technologically advanced Deep Learning algorithms allows for a chance to detect or even prevent a fall, by deploying algorithms that are compact enough to be implemented into a smartphone, but good enough to classify falls from ADL with enough accuracy. In this study, we evaluate the ability of deep learning architecture of relatively small complexity to distinguish between ADL and Fall events, using the combination of CNN and ANN layers for feature extraction and classification respectively. To secure the optimal performance of the model, the algorithm includes the magnitude of the tri-axial acceleration signal (SMV) as is only source of information. The algorithm filters the sensor data windows before they are fed into the model, by only permitting windows that have the value of the last element to be larger than that of the rest of the window. The results show that the model performs well in cases where other models fail, but with relatively lower metrics. After cross-checking with two different datasets, it is confirmed that the model’s performance can generalize for smartphone use.
Description
Keywords
Fall prediction, Deep learning, Accelerometers, Signal processing