Early detection of sit-to-stand transitions in a lower limb orthosis

This paper describes a method for early detection of posture transitions in a powered orthotic brace purposed for assisting elderly people in sit-to-stand transitions. The detection system is constructed using a linear Support Vector Machine classifier that is trained from features extracted from the signals produced by the sensors mounted on the orthosis. The sensors include accelerometers, potentiometers, gyroscopes and force sensors. The data were collected from 9 healthy individuals that performed various activities, including transitions between sitting, standing and walking. The collected sensor data were processed, normalized and manually annotated to label all posture transitions. A Support Vector Machine classifier was trained and validated in a leave-one-out manner to detect early onset of sit-to-stand transitions. Use of lagged windows was examined in efforts to increase accuracy. The proposed method was to detect posture transitions with a high detection rate and a relatively low number of false positives.