METHOD OF IDENTIFICATION AND DEVICES THEREOF

• US 20110282828A1
• Filed: 05/11/2011
• Published: 11/17/2011
• Est. Priority Date: 05/11/2010
• Status: Active Grant

• Current Assignee: Aerial Technologies Inc.
• Original Assignee: McGill University
• Inventors: Frank, Jordan, Precup, Doina, Mannor, Shie
• Granted Patent: US 8,935,195 B2
• Time in Patent Office: Days
• US Class Current: 706/54
• CPC Class Codes: G06V 40/25 Recognition of walking or running movements, e.g. gait recognition

Abstract

Portable wireless devices are ubiquitous in modern society and many of these have integral sensors such as accelerometers, microphones, and Global Positioning Systems (GPS) that can collect data. This creates potential for intelligent applications to recognize the user, or aspects of the user and take appropriate action. According to embodiments of the invention there are presented techniques for representing such time series data which reduce the memory and computational complexity of performing the analysis and classifying the results. The techniques exploit time-delay embedding is to reconstruct the state and dynamics of an unknown dynamical system, Geometric Template Matching to build nonparametric classifiers, and algorithms to address the problem of selecting segments of data from which to build the time-delay models for classification problems.

Claims

1. A method comprising:

receiving at a computer system time series data;

generating with the computer system a time-series data segment comprising a predetermined portion of said time series data;

generating with the computer system time-delay data, the time-delay data being a time-delay embedded representation of the time-series data segment; and

processing with the computer system the time-delay data with a classifier to generate at least a score of a plurality of scores, each score relating to a measure of the match of the time-delay data to a model of a plurality of models.

2. The method according to claim 1 wherein,

generating the time-series data segment comprises applying a noise reduction step.

3. The method according to claim 1 wherein,

the model of the plurality of models is at least one of generated by generating a time-delay embedded representation of a short segment of prior time series data established during a training process.

4. The method according to claim 1 wherein,

the model of the plurality of models is at least one of selected in a directed manner by the classifier, established through geometric template matching, and varied after an iteration of a classification process within the classifier.

5. The method according to claim 1 wherein,

the computer system maintains a buffer of a predetermined number of samples;

and

the classifier generates a new score as new data becomes available.

6. The method according to claim 1 wherein,

the classifier at least one of selects models in a predetermined manner, selectively focuses on data segments requiring classification, operates on different length segments of time series data, and operates on different numbers of repetitions of a pattern within the time series data.

7. The method according to claim 1 wherein,

the classifier employs at least one of temporal smooth, aggregating, majority voting, nearest neighbour

8. The method according to claim 1 wherein,

the model of the plurality of models is established in dependence upon a training process, the training process employing data generated independently from the source of the time series data.

9. The method according to claim 1 wherein,

the generating and processing are performed in real time upon the computer system that forms a predetermined portion of a device, a second predetermined portion of the device either comprising a sensor generating the time series data or an interface circuit for interfacing to a sensor generating the time series data.

10. A non-transitory tangible computer readable medium encoding a computer program, the computer program for execution by a computer system, the computer program relating to a method comprising:

receiving at a computer system time series data;

generating with the computer system a time-series data segment comprising a predetermined portion of said time series data;

generating with the computer system time-delay data, the time-delay data being a time-delay embedded representation of the time-series data segment; and

processing with the computer system the time-delay data with a classifier to generate at least a score of a plurality of scores, each score relating to a measure of the match of the time-delay data to a model of a plurality of models.

11. The non-transitory tangible computer readable medium encoding a computer program according to claim 10 wherein,

generating the time-series data segment comprises applying a noise reduction step.

12. The non-transitory tangible computer readable medium encoding a computer program according to claim 10 wherein,

the model of the plurality of models is at least one of generated by generating a time-delay embedded representation of a short segment of prior time series data established during a training process.

13. The non-transitory tangible computer readable medium encoding a computer program according to claim 10 wherein,

the model of the plurality of models is at least one of selected in a directed manner by the classifier, established through geometric template matching, and varied after an iteration of a classification process within the classifier.

14. The non-transitory tangible computer readable medium encoding a computer program according to claim 10 wherein,

the computer system maintains a buffer of a predetermined number of samples;

and

the classifier generates a new score as new data becomes available.

15. The non-transitory tangible computer readable medium encoding a computer program according to claim 10 wherein,

the classifier at least one of selects models in a predetermined manner, selectively focuses on data segments requiring classification, operates on different length segments of time series data, and operates on different numbers of repetitions of a pattern within the time series data.

16. The non-transitory tangible computer readable medium encoding a computer program according to claim 10 wherein,

the classifier employs at least one of temporal smooth, aggregating, majority voting, nearest neighbour

17. The non-transitory tangible computer readable medium encoding a computer program according to claim 10 wherein,

the model of the plurality of models is established in dependence upon a training process, the training process employing data generated independently from the source of the time series data.

18. The non-transitory tangible computer readable medium encoding a computer program according to claim 10 wherein,

the generating and processing are performed in real time upon the computer system that forms a predetermined portion of a device, a second predetermined portion of the device either comprising a sensor generating the time series data or an interface circuit for interfacing to a sensor generating the time series data.

19. A method comprising:

providing a classifier, the classifier for generating a score relating to a measure of a match of time-delay data to a model of a plurality of models wherein the time-delay data is a time-delay embedded representation of a time series data segment.

20. The method according to claim 19 wherein,

the time series data segment is a predetermined portion of time series data generated by a sensor interfaced to the computer system, and

the model of the plurality of models is at least one of generated by generating a time-delay embedded representation of a short segment of prior time series data established during a training process.