Sparse Representation for Classification of Posture in Bed

Abstract

Redundant dictionaries, also known as frames, offer a non-orthogonal representation of signals, which leads to sparsity in their representative coefficients. As this approach provides many advantageous properties it has been used in various applications such as denoising, robust transmissions, segmentation, quantum theory and others. This paper investigates the possibility of using sparse representation in classification, comparing the achieved results to other commonly used classifiers. The different methods were evaluated in a real-world classification task in which the position of a lying patient has to be deduced based on the data provided by a pressure mattress of 30×11 sensors. The investigated method outperformed most of the commonly used classifiers with accuracy exceeding 92%, while being less demanding on design and implementation complexity.