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.

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.