Chapter 10 - Fall Detection and Motion Classification by Using Decision Tree on Mobile Phone

Abstract

Falls are a common cause of injury for the elderly all over the world. Research indicates that one out of five elderly falls often cause a serious injury. Most are the head injury and hip fracture. The reason of causing hip fracture is falling sideways. In fact, falls are hard to detect. In addition, people carry their mobile phones almost all the time. It would be wonderful if we can defect falls with a mobile phone. Therefore, in this study, a mobile application for fall management, named the Fall Detection System using Mobile Phone (FDSMP), is developed. With such a mobile phone, an elderly fall can be detected immediately almost when the fall occurs. The reaction time, defined as the time period between the occurrence of the fall event and the time when caregivers arrive at the place where the elderly falls can be then shortened. The application will also tell us what type of fall has occurred. In this study, two microelectromechanical systems (MEMSs), i.e., Triaxial accelerometer and gyroscope installed in a mobile are employed to detect falls and a person's daily movements. Then the correlation between the actions of a falling (or a movement) and the signals collected by the two MEMSs are analyzed so as to differentiate whether the person is doing an activity of daily living (ADL) or the user falls. We subdivided fallings into six types. The differentiation and subdivision are performed by using a decision tree. Our experiments show that the accuracy of fall detection and ADL classification is 98.46% and that of the falling classification is 96.57%.