Fall Detection using Accelerometer Calibration

The risk of fall increases in elderly people due to aging factor. In this paper, a new methodology is revised to perform learning and classification of falls using single accelerometer. In learning and classification of falls, the sensor(s) is normally placed at the same location on human body which might not happen practically. In this paper, it is shown that if the sensor is misplaced during classification, the accuracy of fall detection reduces significantly. Furthermore, a calibrated system is designed which detects the falls accurately even if the sensor is misplaced. Classification is performed using Complex Tree, Quadratic Support Vector Machine and Cubic K-Nearest Neighbor, their accuracies are compared in both scenarios; with/without calibration, and the optimum classifier is identified.