A Basic Study on Detection of a sign of Falling Based on Step Width Change Using an Inertial Sensor

Abstract

In this study, detection of walking at different step width was examined as a way to detect a sign of falling, since the step width was considered to be related to the risk of falling. Gait movements were measured with an inertial sensor attached to the lumbar region, and LSTM model, which is a kind of recurrent neural network, was used to detect walking at different step width. The LSTM learned signals during walking of the reference and predicted the signals from the past signals for test walking data at different step width. Walking at different step width was detected by the prediction error. Experimental tests were performed under three walking conditions: healthy person, simulated elderly person, and simulated hemiplegic person. The results suggested that the method can detect walking at different step width, although the effective signals for the detection differed depending on the gait conditions.