Wavelet transforms based fall detection with neuro-fuzzy systems based feature selection

Abstract

This study proposes a method to detect fall with minimum features selected by a non-overlap area distribution measurement (NADM) method. In preprocessing step, wavelet transforms were carried out to extract wavelet coefficients from dataset acquired by subjects. The NADM was used to select the minimum number of features from wavelet coefficients, and then 19 features were finally selected from the 33 features. The performance result of the fall detection was tested with 19 features, and then the sensitivity, accuracy, and specificity were shown to be 95%, 96.13%, and 97.25%, respectively.