Fall Motion Detection with Fall Severity Level Estimation by Mining Kinect 3D Data Stream

This paper proposes an integrative model of fall motion detection and fall severity level estimation. For the fall motion detection, a continuous stream of data representing time sequential frames of fifteen body joint positions was obtained from Kinect’s 3D depth camera. A set of features is then extracted and fed into the designated machine learning model. Compared with existing models that rely on the depth image inputs, the proposed scheme resolves background ambiguity of the human body. The experimental results demonstrated that the proposed fall detection method achieved accuracy of 99.97% with zero false negative and more robust when compared with the state-of-the-art approach using depth of image. Another key novelty of our approach is the framework, called fall severity injury score (FSIS), for determining the severity level of falls as a surrogate for seriousness of injury on three selected risk areas of body: head, hip and knee. The framework is based on two crucial pieces of information from the fall: 1) the velocity of the impact position and 2) the kinetic energy of the fall impact. Our proposed method is beneficial to caregivers, nurses or doctors, in giving first aid/diagnosis/treatment for the subject, especially, in cases where the subject loses consciousness or is unable to respond.