Fall Detection with Bioradar Using Wavelet Analysis and Deep Learning

Bioradiolocation is a remote method for biological object vital signs detection. In particular, it may be used for noncontact fall detection in elderly. The present paper considers the combined usage of wavelet transform and deep learning techniques in fall detection by means of bioradar data processing. The time-frequency representation of the bio-radar signal representing the absolute values of the wavelet transform coefficients was used as input data for convolutional neural network. The network architecture of AlexNet has been adapted to solve the problem of detecting falls. The proposed method for the fall - non-fall classification was tested on the data gathered in realistic surrounding conditions by five volunteers. The proposed method has the accuracy of 95 % and Cohen's kappa of 91 %.