Efficient activity classification from motion inputs

• US 9,398,867 B2
• Filed: 01/06/2014
• Issued: 07/26/2016
• Est. Priority Date: 01/07/2013
• Status: Expired due to Fees

• Current Assignee: Archinoetics LLC
• Original Assignee: Archinoetics LLC
• Inventors: Downs, J. Hunter, Kowahl, Nathan Ronald, Cirit, Fahrettin Olcay, Robinson, William James, Cowan, Richard Moffitt
• Primary Examiner(s): VO, TRUONG V
• Application Number:
  US14/148,678
• Publication Number:
  US 20140195538A1
• Time in Patent Office: 932 Days
• US Class Current: 1/1
• CPC Class Codes: A61B 2562/0219 Inertial sensors, e.g. accelerometers, gyroscopes, tilt switches
• CPC Class Codes: A61B 5/1118 Determining activity level
• CPC Class Codes: A61B 5/1123 Discriminating type of movement, e.g. walking or running A61B5/1116, A61B5/112 take precedence

Abstract

A device worn by a human performing a method using accelerometer data to classify human activities is disclosed. The method uses memory and computation efficiently. A stream of samples is divided into a sequence of sampling periods; each sample has an acceleration value for each axis (e.g., a 3-D accelerometer). Standard deviation of the values for each axis during each sampling period are calculated. A running sum of each axis'"'"' values can be maintained sample-by-sample. Each value is sorted into a bin of a histogram, by quantifying a deviance from a respective mean in standard deviations. A standard deviation produced from samples of a previous period can be used. The histogram is compared with histograms associated with particular activities and a classification output can be produced. Classification outputs from multiple sampling periods can be used for voting. A threshold amount of activity can be required to begin activity classification.

Claims

1. A method of classifying an activity being performed by a moving body, comprising:

receiving a stream of samples of acceleration data from a three-axis accelerometer worn by the moving body, each sample comprising a value for each of the three axes;

establishing a sequence of sampling periods, each having a plurality of samples from the stream;

maintaining a mean of at least a portion of the values for each axis;

sorting the samples received during each of the sampling periods into an activity histogram for that sampling period, the activity histogram comprising a plurality of bins, the sorting comprising

for each of the values of that sample, characterizing a respective difference between that value, and the mean for values of that axis,

and selecting a bin of the plurality of bins of the activity histogram for each sample according to the characterized differences for the values of that sample; and

outputting an activity classification by comparing the activity histogram with activity histograms associated with classified activities.

2. The method of classifying an activity being performed by a moving body of claim 1, wherein the maintaining comprising calculating a respective mean for each of a plurality of sampling periods.

3. The method of classifying an activity being performed by a moving body of claim 1, further comprising producing a repetitions frequency for each axis, and comparing the produced repetitions frequency to stored repetitions frequencies associated with the classified activities.

4. The method of classifying an activity being performed by a moving body of claim 1, wherein the maintaining comprising calculating a respective mean for the values of each axis in a moving window of sampling periods, and using those means for the characterizing of the differences of values in a subsequent sampling period.

5. The method of classifying an activity being performed by a moving body of claim 4, further comprising maintaining a running sum of the values received during a current sampling period and calculating a mean of the values for the current sampling period and updating the mean for the moving window of sampling periods.

6. The method of classifying an activity being performed by a moving body of claim 1, further comprising maintaining a running sum of the values received during each sampling period, discarding the original values after performing the sorting, and using the running sum to calculate respective means for the values of each axis during the sampling period.

7. The method of classifying an activity being performed by a moving body of claim 1, wherein the difference for each value is taken with respect to respective means of the values for each axis received during one or more previous sampling periods.

8. The method of classifying an activity being performed by a moving body of claim 1, further comprising calculating a standard deviation for the values of each axis during each sampling period, and using the standard deviation for the values of one or more previous sampling periods in the characterizing of the differences of the values for each axis.

9. The method of classifying an activity being performed by a moving body of claim 8, further comprising maintaining a running sum of the values received during each sampling period and discarding the original values after performing the sorting.

10. The method of classifying an activity being performed by a moving body of claim 1, further comprising conditioning the performance of one or more portions of the method on detecting a general level of activity in excess of a threshold level.

11. The method of classifying an activity being performed by a moving body of claim 1, wherein the respective mean for the values of each axis is calculated from 1024 samples previous to the samples being received for a current sampling period.

12. The method of classifying an activity being performed by a moving body of claim 1, wherein each sampling period comprises 256 samples.

13. The method of classifying an activity being performed by a moving body of claim 1, wherein a rate at which samples are produced is 100 Hz.

14. The method of classifying an activity being performed by a moving body of claim 1, wherein the method is performed using only integer calculations.

15. The method of classifying an activity being performed by a moving body of claim 1, wherein the sorting of the sampling comprises relatively weighting the values for each axis.

16. The method of classifying an activity being performed by a moving body of claim 15, wherein the three-axis accelerometer comprises a vertical axis, and the value for the vertical axis is weighted more heavily than values for the remaining axes.

17. The method of classifying an activity being performed by a moving body of claim 1, wherein the sorting comprises assigning each value into one of n positions, and the activity histogram has n^3 bins, and each sample is assigned a bin by a weighted combination of the positions for the constituent values of that sample.

18. The method of classifying an activity being performed by a moving body of claim 17, further comprising incrementing a per-bin count and using the counts of each bin as the activity histogram.

19. The method of classifying an activity being performed by a moving body of claim 1, the selecting of the bin for each sample comprises using a respective standard deviation of previously-received values for each axis to classify a current value into one of four categories, a first category being more negative than ⅔

of a standard deviation from the mean, a second category being between the mean and −

⅔

of a standard deviation from the mean, a third category being between the mean and ⅔

of a standard deviation from the mean and a fourth category being farther from the mean than ⅔

of a standard deviation.

20. A device to be worn by a moving body to produce classifications of activities being performed by that creature, comprising:

a multi-axis accelerometer;

a non-transitory memory;

a processor coupled to receive samples, at a sampling rate, of acceleration sensed by the multi-axis accelerometer, each sample comprising a value for each sampled axis, the processor configured to perform a method comprising;

maintaining, for each axis of accelerometer data, a respective current sum and a square of the current sum of the values received during a sampling period for that axis,

estimating a respective standard deviation for the values of each axis during each sampling period;

sorting each sample into one bin of a histogram having n^3 bins, by sorting the per-axis values of that sample into one of n categories, the sorting comprising characterizing an amount of difference between each value of that sample with a mean of previously-received sample values for that axis, the characterizing expressed as a factor of the standard deviation of the values for that axis;

incrementing a per-bin count for each sample sorted into that bin;

comparing the histogram with a set of histograms that have been determined to be indicative of different activities, and

outputting a classification of the activity being performed during the sampling period based on a result of the comparing.

21. The device of claim 20, wherein the estimating of the standard deviation comprises using an integer approximation of a square root of a calculated variance to estimate a standard deviation.

22. The device of claim 21, wherein the method further comprises using the standard deviation for each value to classify that value into one of four categories, the multi-axis accelerometer produces values for each of three axes, and the histogram comprises 64 bins.

23. The device of claim 21, wherein the selecting comprises using the standard deviation for each value to classify that value into one of four categories, a first category being more negative than ⅔

of a standard deviation from the mean, a second category being between the mean and −

⅔

of a standard deviation from the mean, a third category being between the mean and ⅔

of a standard deviation from the mean and a fourth category being farther from the mean than ⅔

of a standard deviation.

24. The device of claim 21, wherein the non-transitory memory is sized to store samples for a single sampling period, and samples for a subsequent sampling period overwrite samples for a previous sampling period in the non-transitory memory when stored.

25. The device of claim 21, wherein the processor is configured to perform the method conditionally upon a generalized level of activity sensed by the accelerometer being greater than a threshold level of activity.

26. The device of claim 21, wherein the processor is configured to perform the method in response to a generalized level of activity sensed by the accelerometer being greater than a threshold level of activity.

27. The device of claim 21, wherein the method which the processor is configured to perform further comprises selecting the set of histograms from available comparison histograms by matching a repetitions per minute calculated for a dominant axis of movement in the accelerometer data with repetitions per minute figures stored with the available comparison histograms.

28. The device of claim 21, wherein the non-transitory memory stores the set of histograms, which are generic among possible users of the device.

29. A non-transitory machine readable medium, the medium storing machine executable instructions for configuring a machine to perform a method of classifying an activity being performed by a moving body, comprising:

receiving a stream of samples of acceleration data from a three-axis accelerometer worn by the moving body, each sample comprising a value for each of the three axes;

establishing a sequence of sampling periods, each having a plurality of samples from the stream;

maintaining a mean of at least a portion of the values for each axis;

sorting the samples received during each of the sampling periods into an activity histogram for that sampling period, the activity histogram comprising a plurality of bins, the sorting comprising

for each of the values of that sample, characterizing a respective difference between that value, and the mean for values of that axis,

and selecting a bin of the plurality of bins of the activity histogram for each sample according to the characterized differences for the values of that sample; and

outputting an activity classification by comparing the activity histogram with activity histograms.