Regularized K-SVD-Based Dictionary Learning Approaches for PIR Sensor-Based Detection of Human Movement Direction

Abstract

The present work shows how a novel solution using regularization concept based K-SVD dictionary learning (DL) can be proposed for passive infrared (PIR) sensor based ambient assisted living (AAL) technologies. In this work, the AAL system focuses on detecting any human movement in specific directions in an unmanned environment. The regularization concept punishes solutions with large values of sparse representation coefficients and this work has successfully implemented regularized K-SVD (RK-SVD) and regularized approximated K-SVD (RAK-SVD) algorithms for a low cost hardware-software based intelligent AAL system, indigenously developed in our laboratory. This work also proposes modified versions of both algorithms (named MRK-SVD and MRAK-SVD algorithms) where novel methods of adapting the regularization parameter have been introduced. Extensive experimentations established that these approaches could significantly improve upon performances of DL based state-of-the-art approaches known, e.g. using multiple cluster pursuit (MCP) approaches, with MRAK-SVD algorithm emerging as the best alternative.