An Energy Efficient and Elastic Edge-Cloud for Computational Sensing in Smart Geriatric Homes

One of the recent transformations in edge computing realm is an emerging computing model known as "edge-cloud computing" which mimics the cloud computing behavior on edge platforms. These hybrid platforms are in demand as they provide privacy preserving computation for always-on sensor nodes at the edge. Conventionally, they have been implemented using off-the-shelf cloud management frameworks in micro data-centers, but at the cost of increased physical volume, power consumption and portability. Alternatively, Single Board Computers (SBCs) have evolved significantly in terms of computing capability and form factor, and consequently, SBC-clusters are now being exploited for the purpose. This paper presents an energy efficient edge-cloud platform that leverages the scalability of low-cost slave clusters along with an architecture tailored for computational sensing in ambient care facilities. For instance, monitoring activities of daily living in a smart elderly care home presents challenges such as variation in occupancy rate and increased demand for time-critical processing/actuation, like detecting falls. This results in several challenges such as sudden increase in data traffic as well as increased energy consumption on the worker nodes. The proposed architecture mitigates this by virtue of computational sensing and efficient task allocation among the slave-clusters. The computing capability of the platform has been demonstrated using an ensemble action recognition classifier. The monthly power budget for the same has been estimated using a behavior model derived from an existing human occupancy data-set. The results illustrate the potential of this platform to aid computational sensing workloads at-scale.