Fuzzy-Based Fine-Grained Human Activity Recognition within Smart Environments

Abstract

With the increasing ageing population, Smart Home (SH) has been under vigorous investigation to enable Ambient Assisted Living (AAL) and foster independent living. Human Activity Recognition (HAR) is the backbone of AAL systems in order to detect Activities of Daily Living (ADL) and provide timely, context-aware assistance. Existing SH based AAL systems primarily focus on coarse-grained activity recognition (AR) and assume successful usage of everyday objects using binary sensors. Limited attention is given to fined-grained AR by verifying the intended object interactions with evidence from multiple heterogeneous sensor data. This paper proposes a fine-grained AR approach which fuses multimodal data from single objects and handles the imprecise nature of non-binary sensor measurements. This approach leverages the fuzzy ontology to model fine-grained actions with imprecise membership states of the sensors in relation to object and fuzzyDL reasoning tool to classify action completion. In addition, a microservice architecture is proposed with a non-intrusive heterogeneous ambient and embedded object based sensing method. The sensing method integrates both off-the-shelf and bespoke devices to collect fine-grained object level interactions. A case study is provided to illustrate the use of the fine-grained AR approach to recognize kitchen-based activities.