Ronnie Smith (cohort student representative)
Stakeholders in AAL include patients (e.g. elderly), their families, caregivers, public (e.g. NHS) and private organisation. Not all of these stakeholders in will have the same primary concerns, but there are two overarching issues ultimately impeding the uptake of AAL systems: acceptability and scalability. Since activity recognition is fundamental to the success of most AAL systems, with a sizeable majority of full-stack AAL systems incorporating it, it is the single largest technological barrier to solving those issues.
Human Activity Recognition (HAR) in the context of Ambient Assisted Living (AAL) has requirements over and above being able to classify activities accurately. Work over the previous two decades has focused on three main approaches, chronologically: data-driven, knowledge-driven, and now hybrid methods that incorporate the best elements of the previous two approaches.
No approach seen thus far has offered a solution that offers adequate coverage of the requirements for human activity recognition in assisted living environments. Indeed there are approaches that do some things rather well under lab conditions, but it is not difficult to surmise that we are a long way off from delivering a solution that would lead to the required levels of adaptability and scalability, without sacrificing all-important user acceptability.
My project investigates how a novel RFID system, where tags are embedded in the floor and environment of an assisted living testbed, can be used to address one or more of the current open challenges in HAR. It is theorised that individuals in the environment can be tracked based on how they disrupt the signals perceived by the antennas in the ceiling, particularly theReceived Signal Strength Indicator(RSSI), Doppler frequency shift, and phase angle rotation measurements. If treated as a machine learning problem, temporal knowledge in the form of a probabilistic representation of the occupants current position can be used in conjunction with semantic reasoning to improve coarse-grained activity recognition.
My working hypothesis is therefore: radio-frequency environmental sensing can be used to detect activities in an assisted living scenario, by tracking both humans and objects in the environment. In particular, it should be possible to maximise the usefulness of this sensing technique by supplanting the existing sensors used in state of the art systems. This will lead to a HAR approach that achieves better overall coverage of requirements that comprise acceptability and scalability.