Tracking RFID groups with spatio-temporal proximity
Systems and methods for determining radio-frequency identification (RFID) tag proximity groups are provided. The method includes receiving RFID tag readings from multiple RFID tags. The method includes determining signal strengths of the RFID tag readings. The method includes determining pairs of RFID tags based on the RFID tag readings. The method also includes implementing a twin recurrent neural network (RNN) to determine proximity groups of RFID tags based on distance similarity over time between each of the pairs of the RFID tags.
1. A method for determining radio-frequency identification (RFID) tag proximity groups, comprising:
receiving RFID tag readings from a plurality of RFID tags;
determining signal strengths of the RFID tag readings;
determining pairs of RFID tags based on the RFID tag readings;
implementing a twin recurrent neural network (RNN) to determine at least one proximity group of RFID tags based on distance similarity over time between each of the pairs of the RFID tags; and
applying an attention mechanism to compare previous trajectory information with new readings and choose trustworthy new readings that are consistent with the previous trajectory information.
2. The method as recited in claim 1 , further comprising:
before feeding the RFID tag readings as input to the twin RNN, dividing a decision window into multiple time slots, wherein each of the multiple time slots has a duration such that each of the plurality of RFID tags is defined as quasi-stationary.
3. The method as recited in claim 1 , further comprising:
applying an attention mechanism to summarize the RFID readings in each time slot to ensure that reliable readings are emphasized while filtering out noisy reports.
4. The method as recited in claim 1 , further comprising:
determining the distance similarity over time based on a reference tag.
5. The method as recited in claim 1 , further comprising:
training the twin RNN towards learning an embedding function ƒ based on a contrastive loss function including both a lower margin (m 1 ) and an upper margin (m 2 )
Loss
(
x
1
,
x
2
,
y
)
=
1
2
y
max
(
0
,
f
(
x
1
)
-
f
(
x
2
)
2
-
m
1
)
2
+
1
2
(
1
-
y
)
max
(
0
,
m
2
-
f
(
x
1
)
-
f
(
x
2
)
2
)
2
where x 1 and x 2 form a pair of the RFID tag reading attributes from two tags and y is a Boolean label indicative of whether the two tags are in a same proximity group.
6. The method as recited in claim 5 , further comprising:
determining an input for the twin RNN based on a linear transformation of a transformed feature x t =ƒ MultiHead (α t ,R t )W m
wherein t represents a current time, a represents a trajectory, R represents readings from a time slot and W m represents a projection matrix.
7. The method as recited in claim 1 , further comprising:
determining the at least one proximity group to include groups of stationary tags and groups of mobile tags.
8. The method as recited in claim 1 , further comprising:
receiving the RFID tag readings from multiple RFID readers placed to allow overlap to eliminate coverage gaps.
9. The method as recited in claim 1 , further comprising:
applying a decision window as an interval in which a proximity between at least one pair of RFID tags is estimated based on a feature vector that includes at least one reading attribute selected from a group including a timestamp, a phase, a Doppler shift and a received signal strength.
10. The method as recited in claim 1 , further comprising:
using the at least one proximity group of RFID tags to determine at least one of:
hidden purchase behaviors of consumers,
popular purchase paths the consumers, and
ongoing billing of the consumers.
11. A computer system for determining radio-frequency identification (RFID) tag proximity groups, comprising:
a processor device operatively coupled to a memory device, the processor device being configured to:
receive RFID tag readings from a plurality of RFID tags;
determine signal strengths of the RFID tag readings;
determine pairs of RFID tags based on the RFID tag readings;
implement a twin recurrent neural network (RNN) to determine at least one proximity group of RFID tags based on distance similarity over time between each of the pairs of the RFID tags; and
apply an attention mechanism to compare previous trajectory information with new readings and choose trustworthy new readings that are consistent with the previous trajectory information.
12. The system as recited in claim 11 , wherein the processor device is further configured to:
before feeding the RFID tag readings as input to the twin RNN, divide a decision window into multiple time slots, wherein each of the multiple time slots has a duration such that each of the plurality of RFID tags is defined as quasi-stationary.
13. The system as recited in claim 11 , wherein the processor device is further configured to:
apply an attention mechanism to summarize the RFID readings in each time slot to ensure that reliable readings are emphasized while filtering out noisy reports.
14. The system as recited in claim 11 , wherein the processor device is further configured to:
determine the distance similarity over time based on a reference tag.
15. The system as recited in claim 11 , wherein the processor device is further configured to:
train the twin RNN towards learning an embedding function ƒ based on a contrastive loss function including both a lower margin (m 1 ) and an upper margin (m 2 )
Loss
(
x
1
,
x
2
,
y
)
=
1
2
y
max
(
0
,
f
(
x
1
)
-
f
(
x
2
)
2
-
m
1
)
2
+
1
2
(
1
-
y
)
max
(
0
,
m
2
-
f
(
x
1
)
-
f
(
x
2
)
2
)
2
where x 1 and x 2 form a pair of the RFID tag reading attributes from two tags and y is a Boolean label indicative of whether the two tags are in a same proximity group.
16. The system as recited in claim 15 , wherein the processor device is further configured to:
determine an input for the twin RNN based on a linear transformation of a transformed feature x t =ƒ MultiHead (α t ,R t )W m
wherein t represents a current time, a represents a trajectory, R represents readings from a time slot and W m represents a projection matrix.
17. The system as recited in claim 11 , wherein the processor device is further configured to:
determine the at least one proximity group to include groups of stationary tags and groups of mobile tags.
18. A computer program product for determining radio-frequency identification (RFID) tag proximity groups, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computing device to cause the computing device to perform the method comprising:
receiving RFID tag readings from a plurality of RFID tags;
determining signal strengths of the RFID tag readings;
determining pairs of RFID tags based on the RFID tag readings;
implementing a twin recurrent neural network (RNN) to determine at least one proximity group of RFID tags based on distance similarity over time between each of the pairs of the RFID tags; and
applying an attention mechanism to compare previous trajectory information with new readings and choose trustworthy new readings that are consistent with the previous trajectory information.