May – end of research
August – end of implementation & testing.
We can assume that
there is a sensor for every square foot of area of the flat / house.
if not, then it is inferred from the times, entrances & exits to
other adjacent areas to the unknown. There should be a minimum of
unknowns since inference will become ambiguous.
The sensors are
not moved and are always working.
Someone will be
periodically reviewing the data whether there is a warning /
emergency or not plus someone to check system viability.
Emergency systems to pick up critical events
would be handled by the specialised system, instant warning. May or
not use same sensors. This system is designed for care
providers to have more information to provide suitable
care rather than be an hour by hour detection / alarm system.
Some conditions if met could produce a warning after 30 mins; some
after 6 hours; some, 2 or more days would create a warning, see
Each house is different, and will have
differing numbers of sensors and qualities.
External doors & windows are higher
priority than internal ones.
The system will work with any number of sensors and the minimum, 1.
This flexibility is essential as people have differing access to
Diagrams are deliver the message quicker than text, visual
- The semantic of the living space is plotted with activity plot overlain.
- Trending can be observed over hours or weeks or months, perhaps its true benefit is on the long term scale.
Life threatening conditions - what level of
support? eg, heart attacks, strokes, falls etc. This system probably wont help with acute medical conditions, more spot the trend of a developing
Conditions which are non-life threatening but
which need care. Dementia, incontinence, onset of deafness /
blindness, mental health issues, and so forth - 3rd
the common problems faced by the elderly & disabled? Who would
know this? Nurses? Care providers. 1st step is to find
out how to get to the people who know. This is to create template
rules to assist the system with the 80% or majority of cases.
generation systems providers.
Methodology - collecting the data.
Divide the flat
into logical units - a schematic will be helpful for this.
unmonitored_outside | entrance1 | kitchen | corridor | living room |
hall | study | bedroom | bathroom | entrance2 | unmonitored_outside
Then show each
unit's typical activities - monitorable states - like tap on / off,
door open / ajar / closed / locked / unlocked
The unit ought to
be thought of as a cell to which there are entry & exit times,
duration occupied, name, list of possible activities, detected
activity and duration, came from, going to.
The unit will have
totals for a single period: total time occupying the cell (hr/day),
total no of entrances, no of exits, activities, total no of change
of states (tap/door)
The above data is collected from the sensors
for a 24 hr period starting at 12 noon / midnight.
Learning the behavioural patterns
is fancy way of saying there are record minimums and maxima, ie most
number of this type of activity. Learning here is taken to mean
finding the parameters of behaviour; for instance, goes to the
toilet between 3 and 6 times a day. but usually 4 times.
abnormal behaviour be dealt with? ie a "spike". This is a
problem. It should be labeled as abnormal and not used in parameters
of normal behaviour.
Once running, it
has a list of max/minimas for each cell; if on a monitoring session,
a record is exceeded for the period, this is logged / warning sent.
shorter periods - say 12hrs could be used as well.
Trending in the
data can be used to track the changes in behaviour, a slope in the
graph, two or more de/increases, significant jumps; these trends can
be used to send warnings, stay in same position (one not designated
as a seating or sleeping area) for more than certain units of time.
1 unit can be 20 mins or 30 mins, adjusted to the necessary factors,
like type of client.
Can have labels
attached to cells with a green tick - "OK" or a circle
with line through "STOP"
Some kind of
knowledge of chronobiology
- science of body clocks, and what is normal for the age of the
client. This can be used as offset.
Templates can be used to help with inference
- some built in rules, if these conditions are met and / or with
certain detected patterns, can be used as a warning. E.G. pacing is
detected & telephone not used for a certain period, can be
symptom of dementia.
Ideas of the “routine” and
non-routine – avoid false positives; a lot of activity in
kitchen, exceeding maxima; too many of these perception that system
is flawed, too many false negatives and system isn't doing its job.
Visitor problem. Differentiation of home
dweller and outside; unlikely to stop owner's cat from being allowed in the residence. Use tags? Best to determine that there are two or more present and set a condition based on that time. And what if there are 2 residents?
Ethical issues involved in collecting data,
assurances to privacy
Like being in front of the doctor. All
personnel involved are vetted. Without webcams cant provide 100%
protection, more false positives, image isn't always need, location
Some study of the
conditions of being disabled /elderly - what are the common
problems? Which things can this system not solve? Which things are
better left to 1st & 2nd generation systems?
Statistical modelling - k-means, EM
algorythim, cluster analysis.