Accurate Fall Detection by Nine-axis IMU Sensor

Fall related injuries are a central problem for the elderly people, therefore many automated fall detectors have been developed. But prevalent methods are neither practical nor poor in accuracy. This paper proposes a novel fall detection algorithm using accelerometers, gyroscopes and magnetometers. In our study, we divide human activities into two categories: lying posture and no-lying posture. We assume that a lying posture is detected after falls. The proposed algorithm has three steps: quaternion Kalman filter, posture recognition, activity intensity analysis. The data is obtained by using nine-axial inertial measurement unit attached on the waist. Using the quaternion Kalman filer the system can obtain body's posture vectors measured in the frame of reference of the ground. The body's posture vectors include Euler angles, quaternion, acceleration. The Euler angles are used to determine the lying posture or no-lying posture. The quaternion and acceleration are used to analyze activity intensity when lying posture are detected. The proposed method features low computational cost and real-time response, in addition has a nice accuracy and convenient in detect falls.