Wavelet Package Analysis based Fall Detection and Diagnosis

Abstract

As a possible hazard to public health, the fall problem has gradually attracted researchers' attention. Detecting the occurrence of fall and diagnosing its intensity can be critical and helpful for deciding the next move. In this work, a novel accelerometer-based signal processing and analysis framework for fall detection and diagnosis is proposed. Wavelet package decomposition (WPD) is introduced for denoising and multi-resolution time-frequency analysis of signals from smartphone's accelerometer sensor. The extraction of features takes into account the statistical properties in both time domain and wavelet domain. To reduce the computational complexity of training and testing the classifiers, the reduction of dimension is performed by additionally evaluating features with principal component analysis (PCA). Then these features become the input of the first classifier for fall detection and if a fall occurs, the second classifier for intensity diagnosis will take the matter further. Test subjects undertake the experiments of falls and activities of daily living (ADLs) to generate data for analysis. The performance of classification based on different algorithms including $k$ -nearest neighbor ( $k$ -NN) and support vector machine (SVM) is presented and compared. The good results indicate this work's applicability in real-world scenarios.