Method for generating a personalized classifier for human motion activities of a mobile or wearable device user with unsupervised learning

1. A method, comprising:

collecting motion activity data from at least one sensor;

applying an initial motion activity classifier function to the motion activity data to produce an initial motion activity posteriorgram;

performing pre-processing and segmenting the motion activity data into windows to produce segmented motion activity data;

extracting sensor specific features from the segmented motion activity data to produce extracted sensor specific features;

generating an updated motion activity classifier function from the extracted sensor specific features;

collecting subsequent motion activity data from the at least one sensor; and

applying the updated motion activity classifier function to the subsequent motion activity data to produce an updated motion activity posteriorgram.

2. The method of claim 1 , further comprising:

storing the collected motion activity data in a training data set; and

wherein the updated motion activity classifier function is generated based upon the training data set.

3. The method of claim 2 , wherein the training data set also includes generalized motion activity data.

4. The method of claim 3 , wherein the initial motion activity classifier function is based upon the generalized motion activity data.

5. The method of claim 2 , further comprising:

storing the collected motion activity data in a training data set; and

wherein the updated motion activity classifier function is generated based upon the training data set.

6. The method of claim 1 , wherein the initial motion activity classifier function is based upon generalized motion activity data.

7. The method of claim 1 , further comprising:

generating a data selection confidence measure after generating the initial motion activity posteriorgram;

wherein the collected motion activity data is stored in a training data set if the data selection confidence measure is less than an upper threshold; and

wherein the collected motion activity data is not stored in the training data set if the data selection confidence measure is greater than the upper threshold.

8. The method of claim 1 , further comprising:

storing the collected motion activity data in a training data set;

determining whether the training data set contains sufficient data prior to updating the motion activity classifier function; and

not updating the motion activity classifier function unless the training data set contains sufficient data.

9. A sensor chip coupled to a system on chip (SOC), the sensor chip comprising:

at least one sensing device; and

a control circuit configured to:

collect motion activity data from at least one sensor of the at least one sensing device;

apply an initial motion activity classifier function to the motion activity data to produce an initial motion activity posteriorgram;

perform pre-processing steps and segmenting the motion activity data into windows to produce segmented motion activity data;

extract sensor specific features from the segmented motion activity data to produce extracted sensor specific features;

generate an updated motion activity classifier function from the extracted sensor specific features;

collect subsequent motion activity data from the at least one sensor of the at least one sensing device; and

apply the updated motion activity classifier function to the subsequent motion activity data to produce an updated motion activity posteriorgram.

10. The sensor chip of claim 9 , wherein the control circuit is further configured to store the collected motion activity data in a training data set; and wherein the updated motion activity classifier function is generated based upon the training data set.

11. The sensor chip of claim 10 , wherein the training data set also includes generalized motion activity data.

12. The sensor chip of claim 11 , wherein the initial motion activity classifier function is based upon the generalized motion activity data.

13. The sensor chip of claim 10 , wherein the control circuit is further configured to store the collected motion activity data in a training data set; and wherein the updated motion activity classifier function is generated based upon the training data set.

14. The sensor chip of claim 9 , wherein the initial motion activity classifier function is based upon generalized motion activity data.

15. The sensor chip of claim 9 , wherein the control circuit is further configured to generate a data selection confidence measure after generating the initial motion activity posteriorgram; wherein the control circuit stores the collected motion activity data in a training data set if the data selection confidence measure is less than an upper threshold, and wherein the control circuit does not store the collected motion activity data in the training data set if the data selection confidence measure is greater than the upper threshold.

16. The sensor chip of claim 9 , wherein the control circuit is further configured to store the collected motion activity data in a training data set, determine whether the training data set contains sufficient data prior to updating the motion activity classifier function, and not determine the motion activity classifier function unless the training data set contains sufficient data.