Apparatus and method for classification of physical orientation

• US 7,684,954 B2
• Filed: 12/31/2007
• Issued: 03/23/2010
• Est. Priority Date: 12/31/2007
• Status: Expired due to Fees

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Shahabdeen, Junaith Ahemed, Gajjala, Anita
• Primary Examiner(s): Lau; Tung S
• Assistant Examiner(s): SUN, XIUQIN
• Application Number:
  US12/006,192
• Publication Number:
  US 20090171615A1
• Time in Patent Office: 813 Days
• Field of Search: 702/141, 702/183
• US Class Current: 702/141
• CPC Class Codes: A61B 2562/0219 Inertial sensors, e.g. accelerometers, gyroscopes, tilt switches
• CPC Class Codes: A61B 5/1126 using a particular sensing technique
• CPC Class Codes: A61B 5/7264 Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems neural networks per se G06N3/00; expert systems per se G06N5/00
• CPC Class Codes: A61B 5/7267 involving training the classification device
• CPC Class Codes: G01C 21/166 Mechanical, construction or arrangement details of inertial navigation systems
• CPC Class Codes: G01C 9/00 Measuring inclination, e.g. by clinometers, by levels
• CPC Class Codes: G16H 50/70 for mining of medical data, e.g. analysing previous cases of other patients

Abstract

A state classifier uses learning obtained from a plurality of training algorithms each adapted to differentiate between states of physical orientation of an object in response to input data from an tri-axial accelerometer. At least two of the training algorithms are trained using data from an accelerometer mounted at a non-ideal angle. The classifier is trained to distinguish between the desired states from data collected from an tri-axial accelerometer device mounted at a plurality of respective angles with respect to a optimal axis on the object, wherein the angles are in the range of −180 degrees to +180 degrees. The classifier may include a plurality of classifiers and a decision fusion module used to combine the decisions from the respective classifiers to ascertain a state.

Claims

1. A method comprising:

training a classifier device using learning obtained from a plurality of training algorithms each adapted to differentiate between states of physical orientation of an object in response to input data from a tri-axial accelerometer;

wherein at least two of the training algorithms are trained using data from the tri-axial accelerometer when the tri-axial accelerometer is mounted on the object at non-ideal mount angles relative to an ideal mount angle having one axis of the tri-axial accelerometer aligned with a gravitational vector; and

wherein the classifier device is trained to distinguish between the states of the object to which the tri-axial accelerometer device is mounted based on data collected from the tri-axial accelerometer device mounted at one or more of a plurality of respective angles with respect to the ideal mount angle on the object, wherein the plurality of respective angles are in the range of −

180 degrees to +180 degrees.

2. A method according to claim 1 wherein the states are sedentary states and the object is a human subject, and the sedentary states are selected from the group:

sitting, standing and lying.

3. A method according to claim 1 wherein the classifier includes a plurality of classifiers each adapted to differentiate between the states of the object under different mount angles of the tri-axial accelerometer.

4. A method according to claim 3 further including using decision fusion to predict a state of the object from classification information received from each of the classifiers.

5. A method according to claim 4 wherein the decision fusion combines the output from each of the classifiers for the prediction.

6. A method according to claim 5 wherein confusion in a decision for each classifier is considered to calculate a confidence level for at least one of the states.

7. A method according to claim 6 wherein the decision fusion produces an output vector including a confidence level for each of the states after combining the decision from the individual classifiers.

8. A method according to claim 1 wherein the non-ideal mount angle is in the range of −

180 degrees to +180 degrees.

9. The method of claim 1, wherein the ideal mount angle and the non-ideal mount angles are independent from the states of physical orientation of the object.

10. An apparatus comprising:

a plurality of classifiers to distinguish between desired states of physical orientation of an object using data collected from an tri-axial accelerometer device mounted on the object, wherein each of the classifiers is configured receive data collected from the tri-axial accelerometer device when mounted on the object, wherein at least one classifier of the plurality of classifiers is trained using first training data and one or more second classifiers are trained using different training data, wherein the different training data correlates to one or more sets of data collected from the tri-axial accelerometer in one or more respective non-ideal mount angles on the object that differ from an ideal mount angle on the object having one axis of the tri-axial accelerometer aligned with a gravitational vector; and

a decision fusion module to identify one of the desired states from classification information received each of the classifiers based on state data collected from the tri-state accelerometer device;

wherein and the ideal mount angle and the at least one non-ideal mount are independent from the desired states of physical orientation of the object.

11. The apparatus according to claim 10 wherein the at least one non-ideal mount angle is in the range of −

180 degrees to +180 degrees.

12. The apparatus according to claim 10 wherein the decision fusion module combines the output from each of the classifiers for the prediction.

13. The apparatus according to claim 12 wherein the decision fusion module considers confusion in a decision for each classifier to calculate a confidence level for at least one of the possible states.

14. The apparatus according to claim 13 wherein the decision fusion module produces an output vector including a confidence level for each of the possible states after combining the decisions from the individual classifiers.

15. The apparatus according to claim 10 wherein the desired states of physical orientation are states of physical orientation of a human subject.

16. The apparatus according to claim 15 wherein the desired states of physical orientation include sedentary states.

17. A method of determining a state of physical orientation of an object, the method comprising:

training a classifier to differentiate between a plurality of possible states of physical orientation of an object based on data collected from a tri-axial accelerometer when mounted on the object at different mount angles, wherein training the classifier comprises;

collecting first data from the tri-axial accelerometer mounted to the object at a first mount angle such that one axis of the tri-axial accelerometer is aligned to a gravitational vector;

collecting one or more second data from the tri-axial accelerometer mounted to the object at one or more second mount angles that differ from the first mount angle;

providing the first data to at least one training algorithm of a plurality of training algorithms to differentiate between possible states of physical orientation of the object;

providing the one or more second data to at least two training algorithms of the plurality of training algorithms to differentiate between the possible states;

generating an output from each of the plurality of training algorithms; and

providing each of the outputs to the classifier to train the classifier to differentiate between the plurality of possible states of the object;

wherein the first mount angle and the one or more second mount angles are independent from the plurality of possible states of physical orientation of the object.

18. The method of claim 17, wherein the output from each of the plurality of training algorithms comprises at least one of a decision tree and a probability distribution.

19. The method of claim 17, wherein the classifier comprises multiple classifiers, and wherein training the classifier comprises:

training each classifier to differentiate between the plurality of possible states of physical orientation of the object based on data collected from a tri-axial accelerometer at a selected mount angle from the first mount angle and the one or more second mount angles; and

combining outputs from each of the multiple classifiers in a decision fusion module to produce a predicted state from the plurality of possible states of physical orientation of the object.

20. The method of claim 19, wherein the predicted state comprises an output vector that includes confidence levels for each of the plurality of possible states.