Automatic exercise segmentation and recognition

1. A physical activity monitoring device, comprising:

a sensor array including one or more sensors configured to measure physical activity attributes of a user while the user is wearing the physical activity monitoring device;

a controller operable to:

receive signal information from the sensor array;

divide the signal information into overlapping segments;

identify predetermined signal characteristics for each overlapping segment;

analyze the predetermined signal characteristics for each overlapping segment using a supervised classifier trained to recognize if the user is actively engaged in a physical activity during the overlapping segment;

automatically determine time intervals where the user is actively engaged in the physical activity using the physical activity attributes; and

automatically determine a type of physical activity the user is actively engaged in during the determined time intervals using the physical activity attributes; and

a reporter to output information regarding the type of physical activity.

2. The monitoring device of claim 1 , where the one or more sensors include an accelerometer.

3. The monitoring device of claim 1 , wherein the signal information includes signals in three dimensions, and wherein the controller is operable to dimensionally reduce signals in two of the three dimensions into a signal in one dimension.

4. The monitoring device of claim 1 , wherein the predetermined signal characteristics include one or more of: a number of autocorrelation peaks, a number of negative autocorrelation peaks, a maximum autocorrelation value, a log of a maximum autocorrelation value, a root-mean-square amplitude, a mean, a standard deviation, a variance, or an integrated root-mean-square amplitude.

5. The monitoring device of claim 1 , wherein the supervised classifier includes a support vector machine, and wherein to analyze the predetermined signal characteristics the controller is operable to:

train the support vector machine with data collected from a plurality of users during time intervals where the users were actively engaged in a physical activity and time intervals where the users were not actively engaged in a physical activity;

generate a set of transformation vectors, a weight vector and a threshold representative of the user actively engaged in a physical activity;

multiply the predetermined signal characteristics by the set of transformation vectors and weight vector to obtain a plurality of multiplication products;

compare the multiplication products to the threshold;

classify a value above the threshold as representative of an overlapping segment wherein the user is actively engaged in a physical activity;

classify a value below the threshold as representative of an overlapping segment wherein the user is not actively engaged in a physical activity; and

classify overlapping segments as being representative of time intervals where the user is likely to be actively engaged in a physical activity based on the classified values.

6. The monitoring device of claim 5 , further comprising an aggregator configured to determine a time interval defined by a plurality of the classified overlapping segments where the user is likely to be actively engaged in a physical activity.

7. The monitoring device of claim 6 , wherein the controller is operable to receive signal information from the sensor array and further operable to receive a set of time intervals from the aggregator, and wherein to determine a type of physical activity the user is actively engaged in during the set of time intervals, the controller is operable to:

divide the signal information into overlapping segments;

identify predetermined signal characteristics for each overlapping segment; and

analyze the predetermined signal characteristics for each overlapping segment using a supervised classifier trained to recognize the type of physical activity the user is actively engaged in during the overlapping segment.

8. The monitoring device of claim 7 , where the supervised classifier includes a support vector machine, and wherein to analyze the predetermined signal characteristics, the controller is further operable to:

train the support vector machine with data collected from a plurality of users during time intervals where the users were engaged in a plurality of types of physical activity;

generate a set of transformation vectors and a weight vector representative of a user engaged in a type of physical activity;

multiply the predetermined signal characteristics by the set of transformation vectors and weight vector to obtain a plurality of multiplication products;

compare the multiplication products to data sets representative of each of a plurality of predetermined activities where the data sets have been predetermined through machine learning; and

classify overlapping segments as representative of a type of physical activity.

9. The monitoring device of claim 7 , further comprising a voting machine configured to determine the type of physical activity the user is likely to be actively engaged in during a time interval where the aggregator has determined that the user is engaged in a physical activity.

10. The monitoring device of claim 1 , wherein the controller is further operable to determine a number of repetitions the user performs of a repetitive physical activity.

11. The monitoring device of claim 10 , wherein the controller is operable to determine the number of repetitions through a counting method that includes:

receiving a signal from the sensor array; and

transforming the signal into a dimensionally reduced signal having at least one fewer dimensions than the signal received from the sensor array;

counting a number of peaks of the dimensionally reduced signal; and

outputting the number of peaks.

12. The monitoring device of claim 11 , wherein the counting method further includes:

determining a set of candidate peaks;

filtering the set of candidate peaks using local period estimates;

filtering the set of candidate peaks using amplitude statistics; and

counting a number of peaks from the set of candidate peaks.

13. The monitoring device of claim 12 , wherein the counting method further includes:

determining a set of candidate valleys;

filtering the set of candidate valleys using local period estimates;

filtering the set of candidate valleys using amplitude statistics;

counting a number of valleys from the set of candidate valleys;

comparing the number of valleys to the number of peaks; and

designating the greater of the number of valleys and the number of peaks as a number of repetitions; and

outputting the number of repetitions.

14. A method of monitoring physical activity, comprising:

measuring, with a sensor array including one or more sensors, physical activity attributes of a user wearing a physical activity monitoring device including the one or more sensors;

automatically determining a set of time intervals where the user is actively engaged in a physical activity based on the physical activity attributes by:

dividing signal information from the sensor array into overlapping segments;

identifying predetermined signal characteristics for each overlapping segment; and

analyzing the predetermined signal characteristics for each overlapping segment using a supervised classifier;

using the supervised classifier to automatically determine a type of physical activity the user is actively engaged in during the determined time intervals based on the physical activity attributes; and

outputting information regarding the type of physical activity.

15. The method of claim 14 , further comprising outputting information regarding a form of the user performing the physical activity.

16. The method of claim 14 , where the supervised classifier includes a support vector machine, and where analyzing the predetermined signal characteristics further includes:

training the support vector machine with data collected from a plurality of users during time intervals where the users were engaged in a plurality of types of physical activity;

generating a set of transformation vectors and a weight vector representative of a user engaged in a type of physical activity;

multiplying the predetermined signal characteristics by the set of transformation vectors and weight vector to obtain a plurality of multiplication products;

comparing the multiplication products to data sets representative of each of a plurality of predetermined activities where the data sets have been predetermined through machine learning; and

classifying overlapping segments as representative of a type of physical activity.

17. The method of claim 14 , further comprising determining the type of physical activity the user is likely to be actively engaged in during a time interval where an aggregator has determined that the user is engaged in a physical activity.

18. A physical activity monitoring device, comprising:

a sensor array including an accelerometer configured to measure physical activity attributes of a user wearing the physical activity monitoring device;

a controller operable to receive acceleration signal information from the sensor array and to automatically determine time intervals where the user is actively engaged in a physical activity by:

dividing the signal information into overlapping segments;

identifying predetermined acceleration characteristics for each overlapping segment; and

analyzing the predetermined acceleration characteristics for each overlapping segment using a supervised classifier trained to recognize if the user is actively engaged in the physical activity during the overlapping segment;

the controller further operable to automatically determine a type of physical activity the user is actively engaged in during the determined time intervals using the physical activity attributes corresponding to the determined time intervals; and

a reporter to output information regarding the type of physical activity.