Applicability of mel-cepstrum in a fall detection system based on infrared depth sensors

A new variant of the mel-cepstrum, based on a different set of filters than those applied in the known variants of the mel-cepstrum, is proposed. Its applicability in the preprocessing of data from infrared depth sensors in a fall detection system is assessed. A set of features, based on the three-dimensional position, velocity and acceleration of the silhouette center of the monitored person, is used as a reference. The performance of Decision Trees and Support Vector Machines, trained using different sets of features, is used as a criterion of comparison.