A Projected OMP-Hybridized Discriminative K-SVD-Based Dictionary Learning Algorithm for Human Activity Recognition From Accelerometer Signals

Human activity and behavior recognition has become part and parcel of remote health monitoring and home automation systems. Our recent works have shown initial success in employing dictionary learning (DL)-based solutions for such systems and K-SVD algorithms and its variants have been proposed and successfully utilized in this regard. Motivated by those, our present work further investigates such problems and shows how a novel DL-based solution using discriminative K-SVD (DK-SVD) algorithm can be proposed to address such human activity recognition problems. In order to strengthen the performance and address more challenging human activities, this article introduces a novel DL approach, named as projected orthogonal matching pursuit-DK-SVD (POMP-DK-SVD), hybridizing characteristic features of POMP and DK-SVD algorithm. The approach, in spirit, proposes to integrate POMP within the sparse coding stage of DK-SVD. The supremacy of the proposed POMP-DK-SVD approach has been firmly established over earlier published K-SVD and modified K-SVD-based approaches, by considering two benchmark human activity recognition problems at hand.