Radar-based Elderly Fall Detection Using Power Spectral Density Features Obtained by Different Methods

Nowadays, it has become increasingly evident that a significant proportion of the elderly population is susceptible to falls. This situation has prompted a shift in the focus of indoor radar studies, with numerous studies now being presented. In this study, radar-based fall detection is performed using conventional machine classification algorithms, in contrast to studies that concentrate on deep learning. The radar data is subjected to preprocessing through the implementation of two distinct approaches. The power spectral density (PSD) of signals obtained from falls and daily activities is calculated using three distinct PSD calculation methods. The PSDs calculated for each method are employed as features in the classification process. The classification of falls and non-falls is determined without error by the Welch method. The effect of other methods on the results is also significant. In addition, the effect of preprocessing approaches is also indicated. It is presented that the impact of the proposed approaches and methods for falls is quite effective.