Fall recognition algorithm for the elderly based on home service robot

Abstract

The elderly living alone may not be able to stand up when they fall, nor can they timely call the police or inform others, so they may face the risk or even life-threatening. Researchers have developed a lot of schemes based on the fall process, but some of these schemes are too difficult to promote because they are inconvenient to wear and some are too limited because of the location of the camera. This paper proposes an accidental fall reorganization method based on the symmetry principle of home service robot. The Kinect can obtain the monitoring video, frame information of the human body is extracted through OpenPose, and four key parameters are used to identify the fall: speed of descent at the center of the articulatio coxae, the human body centerline angle with the ground, and the aspect ratio of rectangle outside human body, and whether the old man has stood. Finally, the home service robot can confirm whether the alarm is needed with the elderly through voice interaction. This method has 97.3% success rate to recognize the fall down behavior.