Fall-Detection Algorithm Using Plantar Pressure and Acceleration Data

Abstract

In this study, experiments are conducted for four types of falls and eight types of activities of daily living with an integrated sensor system that uses both an inertial measurement unit and a plantar-pressure measurement unit and the fall-detection performance is evaluated by analyzing the acquired data with the threshold method and the decision-tree method. In general, the decision-tree method shows better performance than the threshold method, and the fall-detection accuracy increases when the acceleration and center-of-pressure (COP) data are used together, rather than when each data point is used separately. The results show that the fall-detection algorithm that applies both acceleration and COP data to the decision-tree method has a fall-detection accuracy of 95% or higher and a sufficient lead time of 317 ms on average.