Detection of human fall episodes based on coordinates of body tags and numerical differentiation

This paper presents a new method for detection of changes in alignment of the human body, particularly the fall, on the basis of signals acquired from the position sensors placed on the body of the monitored person. The sensors are located on the cuffs, waist and chest. Transformation of data sequence collected from sensors is proposed in order to best distinguish between the collapse from the normal movement. It is based on nonlinear combination of the first two derivatives of the signals being read. Because data from the sensors is sent asynchronously, a numerical algorithm for unevenly sampled data differentiation is proposed. Derivative values are calculated in equidistant nodes through differentiation of a polynomial, which is adjusted by minimizing the mean square error. The developed method can be used in home care telemedicine systems, where it is necessary to long term monitor of multiple vital parameters of people under care.