Behavior detection using RFID in environments with high RFID tag density

1. An apparatus, comprising:

a communication interface; and

processing circuitry to:

receive, via the communication interface, wireless signal data corresponding to a radio frequency identification (RFID) tag, wherein the wireless signal data comprises signal strength data and signal phase data, wherein the signal strength data and the signal phase data correspond to a plurality of wireless signals transmitted by the RFID tag and received by a radio frequency identification (RFID) reader;

generate decomposed signal strength data based on a seasonal decomposition of the signal strength data, wherein the decomposed signal strength data comprises a seasonal data series, a trend data series, and a residual data series;

generate a frequency-phase curve based on the signal phase data, wherein the frequency-phase curve indicates a plurality of signal phases detected across a plurality of transmission frequencies of the plurality of wireless signals;

extract a set of signal strength features based on the decomposed signal strength data;

extract a set of signal phase features based on the frequency-phase curve; and

detect a motion state of the RFID tag using a machine learning classifier, wherein the machine learning classifier is trained to detect the motion state of the RFID tag based at least on the set of signal strength features and the set of signal phase features.

2. The apparatus of claim 1 , wherein:

the RFID tag is associated with a physical asset; and

the processing circuitry is further to detect behavior associated with the physical asset based on the motion state of the RFID tag.

3. The apparatus of claim 2 , wherein:

the physical asset comprises a retail product; and

the behavior associated with the physical asset comprises a human interaction with the retail product.

4. The apparatus of claim 1 , wherein the motion state of the RFID tag comprises:

a time of occurrence of the motion state;

an identifier of the RFID tag;

an indication of whether the RFID tag is stationary or moving; and

a movement energy intensity of the RFID tag.

5. The apparatus of claim 1 , wherein the set of signal strength features comprises:

trend data extracted from the trend data series of the decomposed signal strength data;

residual data extracted from the residual data series of the decomposed signal strength data; and

a standard deviation of received signal strength indicators (RSSIs) extracted from the signal strength data.

6. The apparatus of claim 1 , wherein the set of signal phase features comprises:

a slope of a linear regression of the frequency-phase curve;

a correlation coefficient of a least-squares regression of the frequency-phase curve;

a descending trend corresponding to a phase variation of the plurality of signal phases;

a dynamic range of the plurality of signal phases; and

a number of zero crossings of the frequency-phase curve.

7. The apparatus of claim 1 , wherein the machine learning classifier comprises a voting classifier, wherein the voting classifier is to detect the motion state of the RFID tag based on motion state predictions from a plurality of machine learning models.

8. The apparatus of claim 7 , wherein the plurality of machine learning models comprises a random forest model and a logistic regression model.

9. The apparatus of claim 1 , wherein the machine learning classifier comprises a single model classifier, wherein the single model classifier comprises:

a support-vector machine (SVM) model;

a random forest model; or

a logistic regression model.

10. The apparatus of claim 1 , wherein the processing circuitry is further to:

normalize a sampling rate of the wireless signal data, wherein the sampling rate is to be normalized based on upsampling or downsampling the wireless signal data.

11. The apparatus of claim 1 , wherein the processing circuitry to generate the frequency-phase curve based on the signal phase data is to:

perform spectral linearity tracking on the signal phase data to generate the frequency-phase curve.

12. At least one non-transitory machine accessible storage medium having instructions stored thereon, wherein the instructions, when executed on a machine, cause the machine to:

receive, via a communication interface, wireless signal data corresponding to a radio frequency identification (RFID) tag, wherein the wireless signal data comprises signal strength data and signal phase data, wherein the signal strength data and the signal phase data correspond to a plurality of wireless signals transmitted by the RFID tag and received by a radio frequency identification (RFID) reader;

generate decomposed signal strength data based on a seasonal decomposition of the signal strength data, wherein the decomposed signal strength data comprises a seasonal data series, a trend data series, and a residual data series;

generate a frequency-phase curve based on the signal phase data, wherein the frequency-phase curve indicates a plurality of signal phases detected across a plurality of transmission frequencies of the plurality of wireless signals;

extract a set of signal strength features based on the decomposed signal strength data;

extract a set of signal phase features based on the frequency-phase curve; and

detect a motion state of the RFID tag using a machine learning classifier, wherein the machine learning classifier is trained to detect the motion state of the RFID tag based at least on the set of signal strength features and the set of signal phase features.

13. The storage medium of claim 12 , wherein:

the RFID tag is associated with a physical asset; and

the instructions further cause the machine to detect behavior associated with the physical asset based on the motion state of the RFID tag.

14. The storage medium of claim 13 , wherein:

the physical asset comprises a retail product; and

the behavior associated with the physical asset comprises a human interaction with the retail product.

15. The storage medium of claim 12 , wherein the motion state of the RFID tag comprises:

a time of occurrence of the motion state;

an identifier of the RFID tag;

an indication of whether the RFID tag is stationary or moving; and

a movement energy intensity of the RFID tag.

16. The storage medium of claim 12 , wherein the set of signal strength features comprises:

trend data extracted from the trend data series of the decomposed signal strength data;

residual data extracted from the residual data series of the decomposed signal strength data; and

a standard deviation of received signal strength indicators (RSSIs) extracted from the signal strength data.

17. The storage medium of claim 12 , wherein the set of signal phase features comprises:

a slope of a linear regression of the frequency-phase curve;

a correlation coefficient of a least-squares regression of the frequency-phase curve;

a descending trend corresponding to a phase variation of the plurality of signal phases;

a dynamic range of the plurality of signal phases; and

a number of zero crossings of the frequency-phase curve.

18. The storage medium of claim 12 , wherein the machine learning classifier comprises a voting classifier, wherein the voting classifier is to detect the motion state of the RFID tag based on motion state predictions from a plurality of machine learning models.

19. The storage medium of claim 18 , wherein the plurality of machine learning models comprises a random forest model and a logistic regression model.

20. A method, comprising:

receiving, via a communication interface, wireless signal data corresponding to a radio frequency identification (RFID) tag, wherein the wireless signal data comprises signal strength data and signal phase data, wherein the signal strength data and the signal phase data correspond to a plurality of wireless signals transmitted by the RFID tag and received by a radio frequency identification (RFID) reader;

generating decomposed signal strength data based on a seasonal decomposition of the signal strength data, wherein the decomposed signal strength data comprises a seasonal data series, a trend data series, and a residual data series;

generating a frequency-phase curve based on the signal phase data, wherein the frequency-phase curve indicates a plurality of signal phases detected across a plurality of transmission frequencies of the plurality of wireless signals;

extracting a set of signal strength features based on the decomposed signal strength data;

extracting a set of signal phase features based on the frequency-phase curve; and

detecting a motion state of the RFID tag using a machine learning classifier, wherein the machine learning classifier is trained to detect the motion state of the RFID tag based at least on the set of signal strength features and the set of signal phase features.

21. The method of claim 20 , further comprising:

detecting behavior associated with a physical asset based on the motion state of the RFID tag, wherein the RFID tag is associated with the physical asset.

22. The method of claim 20 , wherein:

the set of signal strength features comprises one or more of:

trend data extracted from the trend data series of the decomposed signal strength data;

residual data extracted from the residual data series of the decomposed signal strength data; or

a standard deviation of received signal strength indicators (RSSIs) extracted from the signal strength data; and

the set of signal phase features comprises one or more of:

a slope of a linear regression of the frequency-phase curve;

a correlation coefficient of a least-squares regression of the frequency-phase curve;

a descending trend corresponding to a phase variation of the plurality of signal phases;

a dynamic range of the plurality of signal phases; or

a number of zero crossings of the frequency-phase curve.

23. A system, comprising:

a radio frequency identification (RFID) reader to communicate with a plurality of radio frequency identification (RFID) tags; and

processing circuitry to:

receive, from the RFID reader, wireless signal data corresponding to a radio frequency identification (RFID) tag of the plurality of RFID tags, wherein the wireless signal data comprises signal strength data and signal phase data, wherein the signal strength data and the signal phase data correspond to a plurality of wireless signals transmitted by the RFID tag and received by the RFID reader;

generate decomposed signal strength data based on a seasonal decomposition of the signal strength data, wherein the decomposed signal strength data comprises a seasonal data series, a trend data series, and a residual data series;

generate a frequency-phase curve based on the signal phase data, wherein the frequency-phase curve indicates a plurality of signal phases detected across a plurality of transmission frequencies of the plurality of wireless signals;

extract a set of signal strength features based on the decomposed signal strength data;

extract a set of signal phase features based on the frequency-phase curve; and

detect a motion state of the RFID tag using a machine learning classifier, wherein the machine learning classifier is trained to detect the motion state of the RFID tag based at least on the set of signal strength features and the set of signal phase features.

24. The system of claim 23 , wherein:

the RFID tag corresponds to a physical asset; and

the processing circuitry is further to detect behavior associated with the physical asset based on the motion state of the RFID tag.

25. The system of claim 23 , wherein:

the set of signal strength features comprises one or more of:

trend data extracted from the trend data series of the decomposed signal strength data;

residual data extracted from the residual data series of the decomposed signal strength data; or

a standard deviation of received signal strength indicators (RSSIs) extracted from the signal strength data; and

the set of signal phase features comprises one or more of:

a slope of a linear regression of the frequency-phase curve;

a correlation coefficient of a least-squares regression of the frequency-phase curve;

a descending trend corresponding to a phase variation of the plurality of signal phases;

a dynamic range of the plurality of signal phases; or

a number of zero crossings of the frequency-phase curve.