A Effective Feature Construction Method for Fall Detection using Smartphone

Abstract

Recent years, smartphone based fall detection solutions have become research hotspots. These previous algorithms always analyze two types of data (accelerometer and gyroscope) and detect fall event on activities of daily life (ADL) of people which does not consider the case on physical exercise, such as, running etc. In this paper, we propose an effective feature construction method to convert a continuously device motion record to a feature vector which can define the occurrence of a fall event accurately. Base on those feature vectors, a heuristic fusion approach is adopted to extract the fall events on ADL with running. Our method runs on four types of refined and unbiased data (Attitude, RotationRate, Gravity and UserAcceleration) providing by iPhone’s Core Motion framework. And 15 volunteers were employed to simulate fall events. The empirical results have demonstrated that the proposed method is effective and reliable on ADL with physical exercise.