Methods and apparatus for anomaly detection in self-checkout retail environments
A system for anomaly detection in a self-checkout environment, comprising a processing unit for receiving transaction data from a self-checkout terminal: characterising an activity based on a set of features extracted from the received transaction data; defining a plurality of active intervals for each characterised activity; determining a meta-feature vector for each defined active interval of the plurality of active intervals; comparing each meta feature vector with a predefined set of vectors; and determining an anomaly based on the comparison.
1. A system for anomaly detection in a self-checkout environment, comprising:
a processing unit configured for:
extracting a set of features from transaction data received from a self-checkout terminal;
characterising an activity based on the set of features;
defining a plurality of active intervals for each characterised activity;
determining a meta-feature vector for each defined active interval of the plurality of active intervals;
comparing each meta-feature vector with a predefined set of vectors;
detecting an anomaly based on the comparison; and
issuing an alert based on detecting the anomaly.
2. The system of claim 1 , wherein for characterising the activity the processing unit is configured for:
extracting a set of features from the received data,
transforming the set of features from the received data into the corresponding set of feature activation values,
determining an evolution of the set of feature activation values over a time interval, and
defining the activity, Activity μ in accordance with:
{ f i ( n )| i ∈featureSet μ ,μ∈[1, K],n∈[N 1 ,N 2 ],N 2 =N 1 +N a },
where N a is a number of frames considered to detect the activity;
f i (n) is an extracted feature set values;
featureSet μ is a set of features which describes the Activity μ ; and
[N 1 , N 2 ] represents a set of consecutive frames extracted in the time interval between a time of an initial frame N1 and a time of a last frame N2.
3. The system of claim 2 , wherein:
the processing unit is further configured for estimating the number of frames defining the activity Na,
the estimation comprises:
defining an activation function A(f i (n),R i ), and
segmenting a time domain into active intervals based on the activation function, and
A(f i (n),R i ) is characterised as
A
(
f
i
(
n
)
,
R
i
)
=
{
1
,
when
R
i
(
f
i
(
n
)
)
is
true
0
,
otherwise
where R i represents a set of rules defined by a user for each feature f i ; f i (n) is active when R i is true.
4. The system of claim 3 , wherein each active interval for all activities is defined as
ACTIVE [N 1 ,N 2 ] μ =∪{n}|∃A ( f i ( n ), R i )=1∀ n∈[N 1 ,N 2 ],μ∈[1, K].
5. The system of claim 4 , wherein the processing unit is further configured for determining a non-active interval between adjacent active intervals.
6. The system of claim 5 , wherein the processing unit is further configured for:
comparing the non-active interval with a threshold, and
combining the adjacent active intervals and the non-active interval into a merged active interval.
7. The system of claim 6 , wherein for determining the meta-feature vector the processing unit is further configured for:
computing a plurality of statistical measures for each feature f i in the active intervals, and
combining the features to form a meta-feature vector for each active interval.
8. The system according to claim 1 , wherein the processing unit is further configured for:
performing a first comparison between the determined meta-feature vector and a first classification system model, and
creating the alert based on the first comparison.
9. The system according to claim 8 , wherein the processing unit is configured for performing a second classification comparison between the determined meta-feature vector and a second classification system model.
10. The system according to claim 9 , wherein the processing unit is configured for:
comparing a performance of the first classification system model and the second classification system model based on the first and second comparison, and
replacing the first classification system model with the second classification system model if a performance of the second classification system model outperforms the performance of the first classification system model.
11. The system according to claim 10 , further comprising a memory configured for storing the first classification system model as a backup model.
12. The system of claim 11 , wherein the processing unit is configured for:
receiving feedback in respect of the alert, and
updating the second classification system model based on the feedback.
13. The system according to claim 11 , wherein the processing unit is configured for:
comparing the performance of the first classification system model with a performance of the backup model, and
replacing the first classification system model with the backup model if the backup model outperforms the first classification system model.
14. A method for anomaly detection in a self-checkout environment, comprising:
extracting a set of features from transaction data received from a self-checkout terminal;
characterising an activity based on the set of features;
defining a plurality of active intervals for each characterised activity;
determining a meta-feature vector for each defined active interval of the plurality of active intervals;
comparing each meta-feature vector with a predefined set of vectors;
detecting an anomaly based on the comparison; and
issuing an alert based on detecting the anomaly.
15. A computer readable programmable medium carrying a computer program stored thereon which when executed by a processing model implements the method according to claim 14 .