COAR: Collaborative and Opportunistic Human Activity Recognition.
Published in DCOSS, 2024
Md Abdullah Al Hafiz Khan, Nirmalya Roy, H M Sajjad Hossain. In Proceeding of the 13th International Conference on Distributed Computing in Sensor Systems, DCOSS-2017. Ottawa, Canada.
Abstract:
The new era of consumer devices ranging from smartphones, smartwatches, and smart jewelries augmented with our everyday activities and lifestyle help postulate human behavior, activity, gesture, social interaction, and gaming experience. Intelligently tasking and sharing the sensing, processing, storing, and computing tasks among those emerging consumer-friendly commodity devices based on their proximities, advocate the development of resource-aware collaborative and opportunistic smart living applications. Motivated by this emerging subsets of phenomenal applications, we first propose a finite-state machine (FSM) based human activity recognition framework which opportunistically exploits the relevant data sources from multiple heterogeneous devices to help infer a variety of user contexts. We depict a lightweight maximum entropy based classifier and exploit the a-priori conditional dependences among the feature sets to opportunistically select the right set of sensors with the most appropriate devices. Experimental results on real data traces demonstrate that our proposed Collaborative Opportunistic Activity Recognition, COAR framework helps infer the activities of daily living with ≈ 90% accuracy.