On-line distinction methods of human falling motions based on machine learning

A hip protector system using an airbag for prevention of femoral neck fractures is under developing by our group. In the system, instance detection of falling motions by using an appropriate on-line algorithm based on sensor signals is required. The purpose of the present paper is to propose online distinction procedures for human falling motions based on the machine learning, such as the support vector machine and the neural network. Three-types of falling motions which cause a femoral neck fracture for elderly people are considered in the present paper. Five distinction procedures for detecting falling motions are proposed in the present paper. In the proposed procedures, one axis gyro sensor and two axis accelerometers are used. The detection performances of the five procedures are evaluated for the three-types of falling motions as well as the many kinds of daily motions. As the results, the procedure based on the neural network considering time series data of sensor signals provides 100% detection rates for the three-types of falling motions. In addition, robust performances of sensor installation errors of the position/angle are evaluated. We confirmed that the proposed method based on the neural network can ensure robust performances for sensor installation errors of the position/angle.