Elderly Fall Detection Using Attention Based dilated CNN and dilated BiLSTM
Vaibhav Soni, Guruwesh Yadav, Vijay Bhaskar Semwal · 2024
The World Health Organisation estimates that thousands of people die each year as a result of falling, which is a serious health issue. When necessary, fall detection and fall prediction tasks allow for precise medical help to be provided to vulnerable populations, which enables local authorities to anticipate daily health care resources and limit fall damages correspondingly. It is important and difficult to identify elderly adults who are at high risk of falling. Wearable sensors have previously been shown to be an effective tool for tracking daily activities. Body-worn sensors like a gyroscope and an accelerometer can offer useful information to input for fall detection. In this paper, we proposed that fall detection activity can be detected by using Dilated CNN and BiLSTM with attention to the KFall and MobiAct datasets and achieve the accuracy of 99.31% and 99.40% in KFall and MobiAct datasets respectively.