Method for identifying a person's posture

• US 9,026,477 B2
• Filed: 02/02/2007
• Issued: 05/05/2015
• Est. Priority Date: 02/02/2006
• Status: Active Grant

• Current Assignee: Commissariat a L'Energie Atomique (French Alternative Energies and Atomic Energy Commission)
• Original Assignee: Commissariat a L'Energie Atomique (French Alternative Energies and Atomic Energy Commission)
• Inventors: Bonnet, Stephane, Godin, Christelle
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Seck, Ababacar
• Application Number:
  US12/162,186
• Publication Number:
  US 20090012921A1
• Time in Patent Office: 3,014 Days
• Field of Search: 706/12
• US Class Current: 706/12
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 2218/12 Classification; Matching
• CPC Class Codes: G06V 30/242 Division of the character sequences into groups prior to recognition; Selection of dictionaries
• CPC Class Codes: G06V 40/23 Recognition of whole body movements, e.g. for sport training

Abstract

A classification method including first classifying an event of any kind by first rules, and then second classifying events, not identified by the first classification, by a learning base reinforced with all the events identified by the first classification. The method is adaptive if the second classification rules are amended according to new examples that were able to be determined by the first rules.

Claims

1. A method for identifying postures of a person, comprising:

attaching a movement sensor on the person, the sensor providing a signal depending on the postures of the person during a time period;

extracting a first signal from the sensor;

making a temporal partition of the first signal into temporal segments each corresponding to a particular one of the postures, said temporal partition;

making a first classification by classifying each temporal segment according to first decision rules, a part of the temporal segments being classified in a class of known posture and another part of the temporal segments, not being classified in a class of known postures, being classified in a class of undetermined posture;

extracting a second signal from the sensor;

creating an automatic learning base, which automatically elaborates second decision rules depending on the second signal and at least some of those temporal segments classified in a class of known postures in the first classification; and

making a second classification by classifying the segments in the class of undetermined posture in a class of known posture of the first classification with said second decision rules,

wherein the making the temporal partition includes using partitioning parameters with an activity index based on at least one of a kinetic energy of the person or a change in the pitch or yaw of the person.

2. The method for identifying postures of a person according to claim 1, further comprising sampling the signal into portions with uniform durations and grouping groups of portions to provide segments.

3. The method for identifying postures of a person according to claim 2, wherein the grouping of groups of portions is carried out according to a criterion of maximum variation of an activity level of the person and a criterion of maximum tilt variation of the person.

4. The method for identifying postures of a person according to claim 1, further comprising:

applying the second decision rules to said segments previously classified into the event classes using the first decision rules.

5. The method for identifying postures of a person according to claim 4, further comprising:

suppressing at least one rule of the first decision rules upon determination that the second decision rules have achieved sufficient reliability.

6. The method for identifying postures of a person according to claim 1, wherein:

the making the temporal partition includes using partitioning parameters with an activity index ( IA) based on average kinetic energy according to a first relation;

7. The method for identifying postures of a person according to claim 1, wherein:

the making the temporal partition includes using partitioning parameters with an activity index (I M) based on yaw activity of the wearer in a medio-lateral axis direction according to a first relation;

I M[n]= h_ML[n+1]−

h_ML[n],

n is an index,

the value h_ML is calculated according to a second relation;

8. A method according to claim 1, wherein the movement sensor includes a first sensor and a second sensor, the first signal being provided by the first sensor and the second signal by the second sensor.

9. A method according to claim 8, wherein the first sensor is an acceleration sensor and the second sensor is a magnetometer.

10. A method according to claim 1, wherein the first decision rules provide a confidence index for each of the temporal segments classified in a class of known posture by the first decision rules, and said temporal segments classified by the first decision rules are exploited by the learning base only if their confidence index is above a predetermined threshold.