Methods, devices, and apparatuses for activity classification using temporal scaling of time-referenced features

Patent metadata

Field	Value
Publication number	US-9407706-B2
Application number	US-201213363147-A
Country	US
Kind code	B2
Filing date	Jan 31, 2012
Priority date	Mar 31, 2011
Publication date	Aug 2, 2016
Grant date	Aug 2, 2016

Abstract

Components, methods, and apparatuses are provided that may be used to obtain a plurality of time-referenced features extracted from signals received from one or more inertial sensors of a mobile device, wherein the time-referenced features are provided according to heterogeneous time scales. The time-referenced features are combined over a decision window to infer an activity of a user co-located with the mobile device.

First claim

What is claimed is:

1. A method comprising:
receiving signals from one or more inertial sensors of a mobile device;
determining, from the received signals, a first time-referenced feature in each of one or more first time windows according to a first time scale and a second time-referenced feature in each of one or more second time windows according to a second time scale, the first time scale different from the second time scale in lengths of the one or more first and second time windows, said first and second time-referenced features each indicating a likelihood of one or more activities using the respective first or second time scale;
combining said first and second time-referenced features over a decision window, wherein a number of the first time window(s) in the decision window is different from a number of the second time window(s) in the decision window; and
inferring an activity of a user co-located with said mobile device based on the combined first and second time-referenced features.

2. The method of claim 1 , wherein combining said first and second time-referenced features over said decision window further comprises:
for each of said first and second time scales, computing activity likelihood function values based, at least in part, on features provided in said first and second time scales over said decision window, said activity likelihood function values being indicative of likelihood functions that said user is performing one of a plurality of activity classifications.

3. The method of claim 2 , wherein, for each of said first and second time scales, said computed activity likelihood function values are combined over said first and second time scales; and
for one or more of said plurality of activity classifications, combining, over said first and second time scales, said activity likelihood function values combined over said decision window.

4. The method of claim 3 , wherein combining, over said first and second time scales, said activity likelihood function values combined over said decision window further comprises weighting combinations of activity likelihood function values obtained over said decision window.

5. The method of claim 4 , wherein combining activity likelihood function values for said first and second time scales over said decision window further comprises applying a weighting factor to a time window of at least one of said first and second time scales.

6. The method of claim 5 , wherein said selecting said activity further comprises applying said combined time-referenced features to a Bayesian classifier to select an activity class from among a plurality of activity classes.

7. The method of claim 1 , wherein said decision window is an integer multiple of at least one of said first and second time scales.

8. The method of claim 1 , wherein at least one of said signals comprises an accelerometer trace.

9. The method of claim 1 , wherein said signals comprise at least three accelerometer traces indicative of acceleration in at least three linear dimensions.

10. The method of claim 1 , further comprising computing joint probabilities using a Gaussian Mixture Model.

11. The method of claim 1 , further comprising:
computing Cepstral coefficients from said signals received from said one or more inertial sensors.

12. The method of claim 1 , further comprising:
computing variance from accelerometer norm.

13. The method of claim 1 , further comprising:
computing accelerometer orientation with respect to gravity.

14. An article comprising:
a non-transitory storage medium comprising machine-readable instructions stored thereon which are executable by a processor of a mobile device and configured to cause the processor to:
receive signals from one or more inertial sensors of a mobile device;
determine, from the signals, a first time-referenced feature in each of one or more first time windows according to a first time scale and a second time-referenced feature in each of one or more second time windows according to a second time scale, the first time scale different from the second time scale in lengths of the one or more first and second time windows, said first and second time-referenced features each indicating a likelihood of one or more activities using the respective first or second time scale;
combine said first and second time-referenced features over a decision window, wherein a number of the first time window(s) in the decision window is different from a number of the second time window(s) in the decision window; and
infer an activity of a user co-located with said mobile device based on the combined first and second time-referenced features.

15. The article of claim 14 , wherein said instructions are further configured to cause the processor to:
for each time scale, compute activity likelihood function values based, at least in part, on features provided in time scales over said decision window, said activity likelihood function values being indicative of likelihood functions that said user is performing one of a plurality of activity classes, and combining said computed activity likelihood function values over said time scale; and
for one or more of said plurality of activity classes, combine, over multiple time scales, said computed activity likelihood function values combined over said decision window to combine said first and second time-referenced features over a decision window.

16. The article of claim 15 , wherein said instructions are further configured to cause the processor to weight combinations of activity likelihood function values obtained over said decision window to combine, over said multiple time scales, said activity likelihood function values combined over said decision window.

17. The article of claim 14 , wherein said non-transitory storage medium further comprises machine-readable instructions stored thereon which are executable by the processor of the mobile device and configured to cause the processor to:
infer said activity by selecting said inferred activity from among a plurality of activity classes.

18. The article of claim 17 , wherein said instructions are further configured to cause the processor to apply said combined time-referenced features to Bayesian classifier to select an activity class from among said plurality of activity classes to select said activity.

19. The article of claim 14 , wherein said time-referenced features are determined using Cepstral filtering.

20. The article of claim 14 , wherein said non-transitory storage medium further comprises machine-readable instructions stored thereon which are executable by the processor of the mobile device and configured to cause the processor to:
compute joint probabilities of said activity likelihood function values using a Gaussian Mixture Model.

21. A mobile device comprising:
one or more inertial sensors; and
one or more processors configured to:
receive signals from one or more inertial sensors of a mobile device;
determine, from the signals, a first time-referenced feature in each of one or more first time windows according to a first time scale and a second time-referenced feature in each of one or more second time windows according to a second time scale, the first time scale different from the second time scale in lengths of the one or more first and second time windows, said first and second time-referenced features each indicating a likelihood of one or more activities using the respective first or second time scale;
combine said first an

Assignees

• Grokop Leonard Henry
• Sarah Anthony
• Nanda Sanjiv
• Qualcomm Inc

Inventors

• Grokop Leonard Henry
• Sarah Anthony
• Nanda Sanjiv

Classifications

• H04L67/22 Primary

  Electricity · mapped topic

• H04L67/125

  involving control of end-device applications over a network · CPC title

• H04M2250/12

  including a sensor for measuring a physical value, e.g. temperature or motion · CPC title

• H04L67/04 Primary

  specially adapted for terminals or networks with limited capabilities; specially adapted for terminal portability · CPC title

• H04M1/72569

  Electricity · mapped topic