Monitoring behaviour

The answers to many questions in the psychological and health sciences require knowing how people are behaving. For example, level of physical activity, smoking, drinking and social contacts are all intimately related to health status. However, existing methods for observing behaviour have been subject to a fundamental problem: people, like subatomic particles, react to being observed. In effect, the act of looking itself influences what is seen (an analogue of the ‘Heisenberg Uncertainty Principle’). People change what they are doing if they know they are being watched and don’t tell the truth when reporting on their own behaviour (e.g., to avoid embarrassing or idiosyncratic activities).

Many researchers have been developing ‘smart home’ systems to monitor everyday activities in a way that largely avoids Heisenberg’s paradox. However, these systems are ad hoc, composed of multiple kinds of devices (e.g., RFID tags, cameras, microphones, infrared presence detectors, temperature/humidity sensors – see Boxlab or FZI Living Lab), complex, fragile, and hence difficult to install or maintain. They often provide a level of detail (e.g., about body orientation) that is not necessary for activity recognition. On the other hand, there are a number of commercially available, relatively simple and robust systems for monitoring activity in home environments. For example, GE’s QuietCare, and Just Checking systems both consist of wireless passive infrared units, together with a base unit that plugs into the wall mains for power, and reports periodically to a website. They can be left in place for months at a time, are unobtrusive, relatively easy to install and cost-effective. However, they are restricted to detecting gross levels of activity (usually by a single person) in given rooms of the house. In effect, they enable macro-scale routine estimation (e.g., time spent in bathroom), not monitoring of behavioural sequences.

My former PhD student, Gaby Judah, and I developed a new kind of smart home system that falls between these two extremes, enabling the robust detection of specific behaviours in natural environments. We adapted the Elpas ‘real time location system’ (RTLS) from Visonic Technologies, normally designed for industrial or institutional use, for use in households. This task is being accomplished together with the indispensable, expert help of technical and sales staff from Visonic Technologies (with whom we have an official partnership) and staff from the CASAS ‘Smart Home’ Project at Washington State University (particularly Allan Drassal and Jim Kusznir).

This RTLS has a number of desirable characteristics:

· integrated: single manufacturer, on a single electrical bus

· robust: shock- and water-proof commercial-grade components (used in industrial contexts such as hospitals and warehouses), redundant communication protocols, system can be left undisturbed for months at a time

· simple: uses only four kinds of components – person/object tags, exciters to define specific detection zones, a Reader for managing the exciters and tags, and a plug computer for storing/transmitting data to the web

· flexible: with a range of sensitivities to inputs, configurations of components, novel types of components, components able to communicate across significant distances

· scalable: additional behaviours can be studied simply by adding tags to the relevant objects

· unobtrusive: components are small in size and blend into the background, with limited wiring

· accurate: low rates of data loss ensure few false negatives; false positives can be managed through use of contextual information

· cost-effective: around £1000 per system, which can be used for many installations

Use of an RTLS is justified on the basis that most behaviours of interest require people to interact with focal objects (e.g., a soap dispenser for handwashing, toothpaste, medicine bottle, kettle, oven, bed, etc.). By having people wear a watch-like sensor, and tagging the objects required to perform each of the target behaviours, we can tell who is doing what, and where and when they are doing it (thanks to other features of the system). Thus, the Visonic system enables us to monitor a wide range of behaviours – including personal hygiene (e.g., handwashing with soap, toothbrushing, flossing, hair-brushing), household cleaning (e.g., dishwashing, surface cleaning), eating-related behaviours, sleeping and medicine-taking – over long periods without disturbing household members. We are currently running field tests to determine the accuracy with which the system can detect the particular events underlying the everyday activities mentioned above. We believe this to be the first use of an integrated RTLS for home-based activity detection.

Our initial plan is to use this system to study habit formation in real time. Previous studies of habit formation have been hampered by requiring study participants to report – typically daily – on their own behaviour, which obviously reinforces the tendency to engage in the target behaviour, and thus significantly influences the very process being studied. For the first time, we will be able to follow the course of performance for months at a time without requiring participants to be constantly reminded that their behaviour is being tracked. This should produce a much more valid study of habit formation than has previously been possible.

By introducing specific features into the environment, or subjecting household members to psychological tests or treatments, we can also measure the long-term consequences of such interventions on everyday behaviour in a way not previously possible. This will include tests of interventions that might improve habit formation.

Because we can cost-effectively monitor many behaviours by the same people on a day-to-day basis simply by adding sensors to more objects around the house, we can measure regularity in the sequencing and timing of daily routines. How people budget their time can illuminate many things about them: how much time do they spend multi-tasking? How often do activities remain incomplete? How quickly can common activities be achieved? How variable are everyday activities from day-to-day? How energetically efficient are people’s routines (i.e., do people do things in the most efficient order, with the least unnecessary movement)?

A related application of this system will be in the field of Pervasive Healthcare (and related fields of m-Health, Ambient Assisted Living, Rehabilitation Monitoring, and Embedded or Ambulatory Assessment). The number of elderly people will double over the next 30-40 years in Europe, halving the ratio of workforce to retirees from 4:1 to 2:1 by 2050. National health budgets are also exploding as a consequence. Significant pressure is therefore on to find ways that help elderly people stay out of institutionalized care situations as long as possible, thus reducing the financial burden on public health systems.

A variety of emergency response systems have been designed to get help to people who have fallen down or had a cardiac arrest at home. However, it is also possible to detect a variety of impending problems with the Visonic RTLS. If an elderly person is living alone, is not aware of any problem, but is nevertheless losing the ability to perform particular kinds of behaviour (such as the so-called ‘Activities of Daily Living’, of bathing, eating, cooking, walking, dressing, household chores, and personal hygiene), output from the RTLS can provide health care workers with a new source of information about people’s needs for support. For example, the system could track trends of eating less, preparing fewer hot meals, not taking medication as regularly, making more frequent and prolonged visits to the bathroom, moving between rooms less often or remaining in bed throughout the day, flagging the need for a home care assistant to provide help with particular activities (such as feeding or bathing). Similar support can be provided to those who are attempting to regain everyday functionality after injury and those who require long-term support due to mental difficulties ranging from retardation to obsessive-compulsive disorder. The RTLS can work as an early warning system to help recognise any emerging problems before they become emergencies – thus preventing or delaying hospital, residential or nursing home admission for as long as possible – or serve as a long-term assistive system that enables those who would otherwise be institutionalized to live independently. An RTLS can also allow people to keep track of personal health projects (e.g., provide support for efforts to quit smoking or exercise more), or simply to document their own daily lives as a kind of personal diary (e.g., various ‘life-logging’ projects).

We believe use of RTLS in home and work environments holds tremendous promise for the scientific study of behaviour in situ, over the long term, cost-effectively, and with minimal interference. It should become the new standard for conducting behavioural studies of all kinds.