Human Posture Recognition Using a Hybrid of Fuzzy Logic and Machine Learning Approaches

Abstract

An autonomous assistive robot needs to recognize the body-limb posture of the person being assisted while he/she is lying in a bed to provide care services such as helping change the posture of the person or carrying him/her from the bed to a wheelchair. This paper presents a data-efficient classification of human postures when lying in a bed using a hybrid fuzzy logic and machine learning approach. The classifier was trained using a relatively small dataset containing 19,800 annotated depth images collected using Kinect from 32 test subjects lying in bed. An overall accuracy of 97.1% was achieved on the dataset. Furthermore, the image dataset including depth and red-green-blue (RGB) images, is available to the research community with the publication of this paper, with the hope that it can benefit other researchers.