Recognition of Human Behavior for Assisted Living Using Dictionary Learning Approach

Abstract

The ambient assisted living (AAL) technology aims to provide more safety and self-sufficiency, while permitting older persons to live self-dependently in their homes. Monitoring of activities of daily living (ADL) is one of the key ideas of AAL. This becomes an interesting research idea in modern world, where condition monitoring of various ADLs and their automatic classification is a big challenge. This paper proposes a new approach for activity recognition of motion primitives relying on the sparse representation of signals, where signals are represented using a sparse combination of atoms from an over-complete dictionary. This paper intends to investigate the suitability of applying dictionary learning algorithms like K-singular value decomposition (K-SVD), which is usually used to construct an over-complete dictionary, for the effective progress of the ADL monitoring system. This paper proposes to formulate the classification approach by using SRC classifiers, based on the dictionaries learned using K-SVD algorithm. We have validated our proposed approach on a publicly available ADL data set of wrist-worn accelerometer sensor for activity recognition. Performance evaluations demonstrate that the proposed method outperforms several other competing methods.