Efficient activity classification from motion inputs

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 n̂3 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.

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, the samples comprising a value for each of the three axes;
• establishing a sequence of sampling periods, each having a plurality of samples from the stream;
• producing a respective mean of the values within each sampling period, for each axis, producing a difference between each value within each sampling period, for each axis and the mean of the values for that axis in that sampling period;
• producing an indication of how much each of the values deviates from the respective mean for the axis to which that value pertains;
• sorting each sample in a sampling period into a location of a histogram according to the indication;
• and
• comparing the resulting histogram with histograms correlated to known activities and producing a classification of the activity based on a result of the comparing.

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 associated with classified activities.

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

• producing a histogram having a plurality of entries, from a stream of samples of acceleration data generated during a sampling period by a three-axis accelerometer worn by the moving body, each sample comprising a value for each of the three axes;
• comparing, by a computer, one or more aspects of the histogram with one or more aspects of a histogram associated with a classified activity;
• and
• outputting a classification of activity being performed during the sampling period, in dependence on the comparing.