Adaptive sensor-based activity classification

• US 8,374,775 B2
• Filed: 11/05/2009
• Issued: 02/12/2013
• Est. Priority Date: 11/05/2009
• Status: Expired due to Fees

• Current Assignee: Apple Inc.
• Original Assignee: Apple Inc.
• Inventors: van der Merwe, Rudolph
• Primary Examiner(s): Zanelli, Michael J.
• Application Number:
  US12/613,438
• Publication Number:
  US 20110106418A1
• Time in Patent Office: 1,195 Days
• Field of Search: None
• US Class Current: 701/408
• CPC Class Codes: G01C 21/12 executed aboard the object being navigated; Dead reckoning
• CPC Class Codes: G06V 20/54 of traffic, e.g. cars on the road, trains or boats

Abstract

An activity state (e.g., stationary or moving) of an intermittently moving mobile device is classified using sensor measurements provided by one or more on-board sensors. The activity state can be determined by a classifier that exploits the separable log-normal distributions of the short-time variance of sensor measurements. The activity state can provide an external non-correlated indication of movement or stationarity to an on-board navigation engine. The on-board navigation engine can use the activity state to improve tracking performance.

Claims

1. A method performed by one or more processors of a mobile device for determining an activity state of the mobile device, the method comprising:

calculating a movement onset threshold based on a statistical model of sensor measurements for a motion sensor of the mobile device, where the statistical model is a stationary mode of a separable log-normal distribution of a short-time variance of the sensor measurements;

receiving a first set of measurements from the motion sensor;

calculating a log-variance signal based on the first set of measurements;

comparing the log-variance signal with the movement onset threshold; and

determining that the log-variance signal exceeds the movement onset threshold.

2. The method of claim 1, further comprising:

calculating a stationarity threshold based on the statistical model;

receiving a second set of measurements from the motion sensor;

updating the log-variance signal based on the second set of measurements;

determining that the updated log-variance signal does not exceed the stationary threshold; and

updating the statistical model using the set of measurements.

3. The method of claim 2, further comprising:

generating an activity state indicating that the mobile device is stationary; and

generating a navigation solution using the activity state.

4. The method of claim 2, where updating the statistical model includes updating a mean and variance of the stationary mode of the separable log-normal distribution.

5. The method of claim 4, where the log-variance signal is given by

6. The method of claim 5, where the short time signal mean values μ

_x, μ

_y, μ

_z are given by

7. The method of claim 6, where the movement onset threshold is given by

T_moving=μ

+α

σ

,

where μ and

σ

are a mean and standard deviation, respectively, of a current estimate of a stationary mode Gaussian probability density function and α

reduces spurious classification.

8. The method of claim 7, where updating the statistical model movement onset threshold is given by

T_stationary=μ

+β

σ

,

where μ and

σ

are a mean and standard deviation, respectively, of a current estimate of a stationary mode Gaussian probability density function and β

reduces spurious classification.

9. The method of claim 8, wherein the mean and variance are updated using the following expressions:

μ

_new=γ

μ

_old+(1−

γ

)l(n),

σ

²_new=γ

σ

²_old+(1−

γ

)(l(n)−

μ

_old)²,

where l(n) is a log-variance and γ

is a weighting factor.

10. The method of claim 1, further comprising:

generating an activity state indicating that the mobile device is moving; and

generating a navigation solution based on the activity state.

11. The method of claim 1, where the measurements are acceleration measurements.

12. A system for determining an activity state of a mobile device, comprising:

a motion sensor;

a processor coupled to the sensor and operable for;

calculating a movement onset threshold based on a statistical model of sensor measurements for a motion sensor of the mobile device, where the statistical model is a stationary mode of a separable log-normal distribution of a short-time variance of the sensor measurements;

receiving a first set of measurements from the motion sensor;

calculating a log-variance signal based on the first set of measurements;

comparing the log-variance signal with the movement onset threshold; and

determining that the log-variance signal exceeds the movement onset threshold.

13. The system of claim 12, where the processor is operable for:

calculating a stationary threshold based on the statistical model;

receiving a second set of measurements from the motion sensor;

updating the log-variance signal based on the second set of measurements;

determining that the updated log-variance signal does not exceed the stationary threshold; and

updating the statistical model using the set of measurements.

14. The system of claim 13, where the processor is operable for:

generating an activity state indicating that the mobile device is stationary; and

generating a navigation solution using the activity state.

15. The system of claim 13, where updating the statistical model includes updating a mean and variance of the stationary mode of the separable log-normal distribution.

16. The system of claim 15, where the log-variance signal is given by

17. The system of claim 16, where the short time signal mean values μ

_x, μ

_y, μ

_z are given by

18. The system of claim 17, where the movement onset threshold is given by

T_moving=μ

+α

σ

,

where μ and

σ

are a mean and standard deviation, respectively, of a current estimate of a stationary mode Gaussian probability density function and a reduces spurious classification.

19. The system of claim 18, where updating the statistical model movement onset threshold is given by

T_stationary=μ

+β

σ

,

where μ and

σ

are a mean and standard deviation, respectively, of a current estimate of a stationary mode Gaussian probability density function and β

reduces spurious classification.

20. The system of claim 19, wherein the mean and variance are updated using the following expressions:

μ

_new=γ

μ

_old+(1−

γ

)l(n),

σ

²_new=γ

σ

²_old+(1−

γ

)(l(n)−

μ

_old)²,

where l(n) is a log-variance and γ

is a weighting factor.

21. The system of claim 12, where the processor is operable for:

generating an activity state indicating that the mobile device is moving; and

generating a navigation solution based on the activity state.

22. The system of claim 12, where the measurements are acceleration measurements.