Rejecting out-of-vocabulary words

• US 8,565,535 B2
• Filed: 08/20/2008
• Issued: 10/22/2013
• Est. Priority Date: 08/20/2007
• Status: Active Grant

• Current Assignee: Qualcomm, Inc.
• Original Assignee: Qualcomm, Inc.
• Inventors: Shamaie, Atid
• Primary Examiner(s): ZARKA, DAVID PETER
• Application Number:
  US12/194,780
• Publication Number:
  US 20090052785A1
• Time in Patent Office: 1,889 Days
• Field of Search: 382/209
• US Class Current: 382/209
• CPC Class Codes: G06F 3/017 Gesture based interaction, e.g. based on a set of recognized hand gestures interaction based on gestures traced on a digitiser G06F3/04883
• CPC Class Codes: G06F 3/0346 with detection of the device orientation or free movement in a 3D space, e.g. 3D mice, 6-DOF [six degrees of freedom] pointers using gyroscopes, accelerometers or tilt-sensors
• CPC Class Codes: G06F 3/0488 using a touch-screen or digitiser, e.g. input of commands through traced gestures
• CPC Class Codes: G06V 40/28 Recognition of hand or arm movements, e.g. recognition of deaf sign language static hand signs G06V40/113

Abstract

Enhanced rejection of out-of-vocabulary words, in which, based on applying an input gesture to hidden Markov models collectively modeling a vocabulary of training gestures, a likelihood that the input gesture matches each training gesture, and a quantity of states of the input gesture that match corresponding states of a modeled training gesture determined to have a highest likelihood are determined. The input gesture is rejected if the determined quantity does not satisfy a threshold.

Claims

1. A computer-implemented method comprising:

determining, based on applying an input gesture to a vocabulary of modeled training gestures;

a likelihood that the input gesture matches each modeled training gesture, and

a modeled training gesture having a highest likelihood;

subsequent to determining the modeled training gesture having the highest likelihood, determining a quantity of states of the input gesture that match corresponding states of the modeled training gesture determined to have the highest likelihood, wherein each state of both the input gesture and the modeled training gesture comprises a portion of a gesture; and

rejecting, with a processor, the input gesture if the determined quantity of matching states does not satisfy a threshold, wherein the threshold expresses a quantity or percentage of the total quantity of the states of the input gesture.

2. The method of claim 1, further comprising recognizing the input gesture if the determined quantity satisfies the threshold.

3. The method of claim 2, further comprising controlling an application based on the recognized input gesture.

4. The method of claim 2, further comprising determining an input command associated with the modeled training gesture determined to have the highest likelihood if the determined quantity satisfies the threshold.

5. The method of claim 1, further comprising:

determining a second quantity of states of the rejected input gesture that match corresponding states of a second modeled training gesture determined to have a second highest likelihood; and

rejecting the rejected input gesture if the determined second quantity fails to satisfy the threshold.

6. The method of claim 1, further comprising:

applying observation symbols defining the input gesture to hidden Markov models.

7. The method of claim 1, further comprising:

rejecting the input gesture if a sequence of states of the input gesture does not match a sequence of the corresponding states of the modeled training gesture determined to have the highest likelihood.

8. The method of claim 1, wherein the likelihood and a sequence of state changes are determined using the Viterbi algorithm.

9. The method of claim 1, wherein, if the determined quantity fails to satisfy the threshold, the input gesture is rejected as being outside of the vocabulary.

10. The method of claim 1, further comprising:

training hidden Markov models to recognize the vocabulary of training gestures.

11. The method of claim 1, further comprising:

extracting a set of state changes associated with the input gesture,

wherein the quantity of states of the input gesture are determined from the extracted set of state changes.

12. The method of claim 1, wherein the threshold is expressed as a quantity of the corresponding states.

13. The method of claim 1, further comprising normalizing the input gesture.

14. A computer-implemented method comprising:

determining, based on applying an input gesture to a vocabulary of modeled training gestures;

a likelihood that the input gesture matches each training gesture, and

a quantity of states of the input gesture that match corresponding states of a modeled training gesture determined to have a highest likelihood; and

rejecting, with a processor, the input gesture if the determined quantity does not satisfy a threshold;

wherein determining the quantity of states of the input gesture that match corresponding states of the modeled training gesture determined to have the highest likelihood further comprises;

determining the quantity of states for which an extracted median for the input gesture in each state is greater than or equal to a minimum of extracted medians in a corresponding state for a set of training samples of the modeled training gesture determined to have the highest likelihood.

15. The method of claim 14, wherein determining the quantity of states of the input gesture that match corresponding states of the modeled training gesture determined to have the highest likelihood further comprises:

determining the quantity of states for which an extracted median for the input gesture in each state is greater or equal, by a non-zero adjustment parameter c, to a minimum of extracted medians in a corresponding state for a set of training samples of the modeled training gesture determined to have the highest likelihood.

16. A device comprising a processor configured to:

determine, based on applying an input gesture to a vocabulary of modeled training gestures;

a likelihood that the input gesture matches each modeled training gesture, and

a modeled training gesture having a highest likelihood;

subsequent to determining the modeled training gesture having the highest likelihood, determine a quantity of states of the input gesture that match corresponding states of the modeled training gesture determined to have the highest likelihood, wherein each state of both the input gesture and the modeled training gesture comprises a portion of a gesture; and

reject the input gesture if the determined quantity of matching states fails to satisfy a threshold, wherein the threshold expresses a quantity or percentage of the total quantity of the states of the input gesture.

17. The device of claim 16, further comprising:

an input module configured to receive the input gesture.

18. The device of claim 17, wherein the input module is an interactive screen comprising a camera.

19. The device of claim 16, wherein the likelihood and a sequence of state changes are determined using the Viterbi algorithm.

20. A non-transitory computer-readable medium encoded with a computer program comprising instructions that, when executed, operate to cause a computer to perform operations comprising:

determining, based on applying an input gesture to a vocabulary of modeled training gestures;

a likelihood that the input gesture matches each modeled training gesture, and

a modeled training gesture having a highest likelihood;

subsequent to determining the modeled training gesture having the highest likelihood, determining a quantity of states of the input gesture that match corresponding states of a modeled training gesture determined to have the highest likelihood, wherein each state of both the input gesture and the modeled training gesture comprises a portion of a gesture; and

rejecting the input gesture if the determined quantity of matching states fails to satisfy a threshold, wherein the threshold expresses a quantity or percentage of the total quantity of the states of the input gesture.

21. An apparatus comprising:

means for determining, based on applying an input gesture to a vocabulary of modeled training gestures;

a likelihood that the input gesture matches each modeled training gesture, and

a modeled training gesture having a highest likelihood;

means for determining, subsequent to determining the modeled training gesture having the highest likelihood, a quantity of states of the input gesture that match corresponding states of the modeled training gesture determined to have the highest likelihood, wherein each state of both the input gesture and the modeled training gesture comprises a portion of a gesture; and

means for rejecting the input gesture if the determined quantity of matching states does not satisfy a threshold, wherein the threshold expresses a quantity or percentage of the total quantity of the states of the input gesture.

22. The apparatus of claim 21, further comprising means for recognizing the input gesture if the determined quantity satisfies the threshold.

23. The apparatus of claim 21, wherein, if the determined quantity fails to satisfy the threshold, the input gesture is rejected as being outside of the vocabulary.

24. The apparatus of claim 21, wherein the means for determining the quantity of states of the input gesture that match corresponding states of the modeled training gesture determined to have the highest likelihood further comprises:

means for determining the quantity of states for which an extracted median for the input gesture in each state is greater than or equal to a minimum of extracted medians in a corresponding state for a set of training samples of the modeled training gesture determined to have the highest likelihood.