Falls are serious problems that cause injury and death. The objective of this study was to investigate appropriate Artificial Neural Network (ANN) structures for fall detection using accelerometer and gyroscope sensors. Thresholding techniques and ANN were used to detect falls. In experimental setting, participants were asked to have accelerometer and gyroscope sensors on their waist. The training and testing datasets including falling, standing, sitting, running, and jumping were collected and analyzed. ANN model with feed-forward backpropagation and multi-layer perceptron was tested. Evaluation Indices were used to identify the number of neurons (1 to 30) in the hidden layer. The results indicated that in the hidden layer, there should be a set of 13 neurons, attaining the values of accuracy, sensitivity and specificity as high as 1 each. Therefore, a new fall detection system using wearable sensors can be developed in the future.
Keywords
Evaluation of Artificial Neural Networks Structures for Fall Detection Using Wearable Sensors
THE JOURNAL OF KMUTNB
Published by : King Mongkut's University of Technology North Bangkok Contributions welcome at : http://www.journal.kmutnb.ac.th
By using our website, you acknowledge that you have read and understand our Cookie Policy and Privacy Policy.