Apparatus, system, and method for detecting activities and anomalies in time series data

Patent metadata

Field	Value
Publication number	US-9833196-B2
Application number	US-201214002661-A
Country	US
Kind code	B2
Filing date	Mar 1, 2012
Priority date	Mar 2, 2011
Publication date	Dec 5, 2017
Grant date	Dec 5, 2017

Abstract

Activities and abnormalities in activities are detected by: (1) receiving data corresponding to measurements of an activity occurring during a time interval; (2) determining a plurality of primitives associated with the data, wherein each of the plurality of primitives represents a characteristic pattern in a portion of the time interval; (3) derive an activity structure relating a first subset of the plurality of primitives that are correlated in time; and (4) based on the activity structure, classify a second subset of the plurality of primitives as an abnormal instance of the bodily activity.

First claim

What is claimed is:

1. A non-transitory computer-readable storage medium, comprising executable instructions to:
receive time series sensor data from at least one body-worn sensor of a wearable device, the time series sensor data corresponding to measurements of a bodily activity occurring during a time interval;
determine from the time series sensor data in the time interval at least one stable sensor value;
identify stable regions in the time interval, each stable region corresponding to a portion of the time interval in which values of the time series sensor data remain within a predefined variation range around one of the at least one stable sensor value, the predefined variation range associated with a sensor type;
identify subsequences in the time series sensor data, each subsequence comprising a pattern of the time series sensor data between two sequential stable regions;
identify at least two activity primitives from the subsequences, each activity primitive representing a recurrent pattern of the time series sensor data;
derive an activity structure relating a first subset of the at least two activity primitives, the activity structure representing a normal instance of the bodily activity;
based on the activity structure, classify a second subset of the at least two activity primitives as an abnormal instance of the bodily activity; and
generate, by the wearable device, a signal to an output component of the wearable device for an audible or tactile alert in response to identifying that the abnormal instance of the bodily activity has occurred.

2. The non-transitory computer-readable storage medium of claim 1 , wherein the executable instructions to receive the time series sensor data include executable instructions to:
receive a multi-dimensional time series corresponding to the measurements of the bodily activity collected from multiple sensors, and
wherein each of the at least two activity primitives represents a single-dimensional recurrent pattern in the multi-dimensional time series.

3. The non-transitory computer-readable storage medium of claim 1 , wherein the at least one stable sensor value comprises a speed value or an acceleration value.

4. The non-transitory computer-readable storage medium of claim 1 , wherein the executable instructions to derive the activity structure include executable instructions to:
classify the first subset as a normal instance of the bodily activity if the first subset has at least a minimum degree of temporal overlap.

5. The non-transitory computer-readable storage medium of claim 1 , wherein the executable instructions to classify the second subset include executable instructions to:
compare the second subset to the first subset; and
classify the second subset as the abnormal instance of the bodily activity if the second subset omits at least one of the activity primitives included in the first subset.

6. The non-transitory computer-readable storage medium of claim 5 , wherein the executable instructions to classify the second subset further include executable instructions to:
classify the second subset as the abnormal instance of the bodily activity if the second subset includes at least a minimum number of activity primitives included in the first subset.

7. The non-transitory computer-readable storage medium of claim 1 , wherein the time series sensor data correspond to measurements of a bodily movement, and further comprising executable instructions to:
generate the alert of the abnormal instance of the bodily movement.

8. The non-transitory computer-readable storage medium of claim 1 , wherein the time series sensor data correspond to measurements of a bodily function, and further comprising executable instructions to:
generate the alert of the abnormal instance of the bodily function.

9. A non-transitory computer-readable storage medium, comprising executable instructions to configure a processor to monitor activities of a subject by monitoring data from a plurality of sensors of a wearable device, the executable instructions including instructions to:
receive time series data corresponding to sensor measurements of the plurality of sensors of the wearable device, the time series data including components in a plurality of dimensions;
determine a plurality of subsequences associated with the time series data, wherein each of the plurality of subsequences represents a pattern projected along one of the plurality of dimensions, the pattern being between two sequential stable regions along the one of the plurality of dimensions;
identify a first subset of the plurality of subsequences as representing a normal instance of a first bodily activity based on a degree of temporal overlap of members of the first subset;
identify a second subset of the plurality of subsequences as an abnormal instance of the first bodily activity, wherein identifying the second subset comprises comparing the second subset to the first subset, and classifying the second subset as the abnormal instance of the first bodily activity that comprises determining that the second subset omits at least one subsequence included in the first subset and determining that the second subset includes at least a minimum number of subsequences included in the first subset; and
generate, by the wearable device, an indication of the abnormal instance of the first bodily activity, to direct an output component of the wearable device to generate an audible or tactile alert.

10. The non-transitory computer-readable storage medium of claim 9 , wherein the two stable regions are two of a plurality of stable regions, and each of the plurality of stable regions is a portion of the time series data in which a value of the time series data is within one of a plurality of predefined variation ranges.

11. The non-transitory computer-readable storage medium of claim 9 , wherein the executable instructions to identify the first subset include executable instructions to:
perform graph clustering on vertices corresponding to the plurality of subsequences.

12. The non-transitory computer-readable storage medium of claim 11 , wherein the executable instructions to perform graph clustering include executable instructions to:
assign weights to the vertices based on a degree of temporal overlap of respective pairs of the vertices.

13. The non-transitory computer-readable storage medium of claim 9 , further comprising:
identify a third subset of the plurality of subsequences as a normal instance of a second bodily activity based on a degree of temporal overlap of members of the third subset, wherein the first bodily activity is distinct from the second bodily activity.

14. The non-transitory computer-readable storage medium of claim 13 , further comprising:
identify a fourth subset of the plurality of subsequences as an abnormal instance of the second bodily activity based on comparing the fourth subset to the third subset; and
produce an indication of the abnormal instance of the second bodily activity.

15. A system for abnormality detection, comprising:
a processing unit; and
a memory connected to the processing unit and including executable instructions to configure the processing unit to:
receive time series data from a plurality of body-worn sensors of a wearable device, the time series data of the plurality of body-worn sensors together forming a multi-dimensional time series, the multi-dimensional time series corresponding to measurements of a bodily activity;
determine for each dimension of the multi-dimensional time series a plurality of primitives, each primitive occurring between t

Assignees

• Sarrafzadeh Majid
• Vahdatpour Alireza
• Univ California

Inventors

• Sarrafzadeh Majid
• Vahdatpour Alireza

Classifications

• A61B5/7271 Primary

  Specific aspects of physiological measurement analysis (specific diagnostics methods using bioelectric or biomagnetic signals A61B5/316) · CPC title

• A61B5/7246

  using correlation, e.g. template matching or determination of similarity · CPC title

• A61B5/11

  Measuring movement of the entire body or parts thereof, e.g. head or hand tremor or mobility of a limb {(A61B5/1038 takes precedence; motion detection to correct for motion artifacts in physiological signals A61B5/721)} · CPC title

• G16H50/20

  for computer-aided diagnosis, e.g. based on medical expert systems · CPC title

• A61B5/1118 Primary

  Determining activity level · CPC title