Development of a Health-Monitoring Device for Activity Recognition and Fall Detection

Abstract

Activity recognition plays a crucial role in health monitoring systems. Most of our vital parameters like heartbeat rate or blood pressure are dependent on the activity we are doing at the time, and without knowing that, it is hard to figure out the anomalies. Besides, activity recognition can help us to detect emergency situations like falling, or even heart stroke. Knowing the importance of detecting unusual activities (, e.g. falling) and usual activities (, e.g. walking), this research has investigated the possibility of detecting fall and Activities of Daily Living (ADLs) by the help of the three dominant frequencies of accelerations of wrist in each axis and their amplitudes. In this regard, a wearable device is designed with an IMU to detect walking, running, staying still and falling. Decision function (statistical model) is calculated using ANN. To train the function, 674 samples are gathered from almost 30 people. Results show 94.8% accuracy in detecting ongoing activity and if we only consider distinguishing fall from ADLs, the values for accuracy, sensitivity and specificity are 96%, 88% and 98%, respectively.