132
D. Zekri et al.
Our notion of activity comprises two key criteria used also in [10] that are
at the basis of our verification process:
1. Location: the specific place where an activity occurs, for example, “eating”
takes place in the kitchen.
2. Time: the duration and occurring time of an activity. The user may perform
a same activity at different times (e.g., going to the toilet) but some activities
only occur at specific times of the day (e.g., eating breakfast). The start time
and duration of each activity instance may be logged by the user, or better
detected by an activity recognition system based on in-home sensors.
Let A = {a 1 ; a 2 ; ...; a 4 } the set of activities labels. An activity pattern represents when and where an activity usually occurs. It is defined as a tuple:
P a = (a i , S a (t), D a (t))
where:
– a i ∈ A is an activity label
– S a (t) is a time interval representing the usual start time of activity a i
– D a (t) is a time interval representing the usual duration of activity a i
The daily behavior pattern involves several activity patterns. It defines order
constraints on them and introduces eventual temporal delays. The daily behavior
pattern describes how the user performs her/his activities at different times
and models links between them. The daily behavior pattern is represented by a
sequence of usual activities. It can be built from data derived from sensors in a
smart home.
B = (P a1 , P a2 , P a3 ) Where P ai is an activity pattern
For each day of the week D i we built a behavior pattern B i which is a set
of segments P ai , where each segment P ai is a sequence of tuples a i , S a (t), D a (t)
related to each activity. In this pattern we consider three activities: “sleeping”,
“eating”, “taking a shower”, which occur at specific times of the day. “Going to
the toilet” may occur at many times during the day. It will be studied separately
as we will see later.
3.2 Normal Behavior Pattern for the Elderly
The first step of any behavior anomaly detection system is to characterize the
normal behavior, also called routine behavior or regular behavior, based on training data to model regularities in every individual activity. The normal behavior
consists of the list of activities that a resident performs in her/his house, with
time of the day and the duration. Thus, it captures the repetitive daily routines
and deviations from the normal behavior may indicate changes of lifestyle or loss
of capacity.
D. Zekri et al.
Our notion of activity comprises two key criteria used also in [10] that are
at the basis of our verification process:
1. Location: the specific place where an activity occurs, for example, “eating”
takes place in the kitchen.
2. Time: the duration and occurring time of an activity. The user may perform
a same activity at different times (e.g., going to the toilet) but some activities
only occur at specific times of the day (e.g., eating breakfast). The start time
and duration of each activity instance may be logged by the user, or better
detected by an activity recognition system based on in-home sensors.
Let A = {a 1 ; a 2 ; ...; a 4 } the set of activities labels. An activity pattern represents when and where an activity usually occurs. It is defined as a tuple:
P a = (a i , S a (t), D a (t))
where:
– a i ∈ A is an activity label
– S a (t) is a time interval representing the usual start time of activity a i
– D a (t) is a time interval representing the usual duration of activity a i
The daily behavior pattern involves several activity patterns. It defines order
constraints on them and introduces eventual temporal delays. The daily behavior
pattern describes how the user performs her/his activities at different times
and models links between them. The daily behavior pattern is represented by a
sequence of usual activities. It can be built from data derived from sensors in a
smart home.
B = (P a1 , P a2 , P a3 ) Where P ai is an activity pattern
For each day of the week D i we built a behavior pattern B i which is a set
of segments P ai , where each segment P ai is a sequence of tuples a i , S a (t), D a (t)
related to each activity. In this pattern we consider three activities: “sleeping”,
“eating”, “taking a shower”, which occur at specific times of the day. “Going to
the toilet” may occur at many times during the day. It will be studied separately
as we will see later.
3.2 Normal Behavior Pattern for the Elderly
The first step of any behavior anomaly detection system is to characterize the
normal behavior, also called routine behavior or regular behavior, based on training data to model regularities in every individual activity. The normal behavior
consists of the list of activities that a resident performs in her/his house, with
time of the day and the duration. Thus, it captures the repetitive daily routines
and deviations from the normal behavior may indicate changes of lifestyle or loss
of capacity.