Development of an awakening behavior detection system using a neural network
Hironobu Satoh, Fumiaki Takeda, Yuhki Shiraishi, Rie Ikeda · Electronics and Communications in Japan · 2011
Abstract We have developed a behavior detection system using a neural network (NN). The system detects dangerous behaviors such as nearly falling out of bed and actual falling out of bed. For detection, the system uses pictures captured by a web camera. The system classifies subjects' behavior into five states using a NN. The five states are “lying in bed,” “starting to sit up in bed,” “sitting in bed,” “almost falling from the bed,” and “having fallen from the bed.” Then, we define states of almost falling and having fallen from the bed as dangerous behavior. The states of lying in bed, starting to sit up in bed, and sitting in bed are defined as safe behavior. We propose a final detection rule by which the system classifies their behavior into two states. Finally, the detection ability of the system is evaluated. From the experiment's results, the detection rate of dangerous behavior is shown to be 81.7% using the proposed system. © 2011 Wiley Periodicals, Inc. Electron Comm Jpn, 94(2): 42–50, 2011; Published online in Wiley Online Library ( wileyonlinelibrary.com ). DOI 10.1002/ecj.10296