Rule-based combination of a hierarchy of classifiers for occlusion detection

• US 8,331,674 B2
• Filed: 04/06/2007
• Issued: 12/11/2012
• Est. Priority Date: 04/06/2007
• Status: Active Grant

• Current Assignee: International Business Machines Corporation, International Business Machines SA (International Business Machines Corporation)
• Original Assignee: International Business Machines Corporation
• Inventors: Feris, Rogerio, Hampapur, Arun, Tian, Ying-Li
• Primary Examiner(s): Akhavannik, Hadi
• Application Number:
  US11/697,535
• Publication Number:
  US 20080247609A1
• Time in Patent Office: 2,076 Days
• Field of Search: 382/103, 382/181
• US Class Current: 382/181
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition
• CPC Class Codes: G06V 40/161 Detection; Localisation; Normalisation

Abstract

An occlusion detection system and method include a decomposer configured to decompose an image into a set of hierarchical parts. A hierarchy of classifiers is employed to detect features in the image and the hierarchical parts. A logical operation is configured to logically combine a classification result from at least two of the classifiers to detect an occlusion state of the image.

Claims

1. An occlusion detection system, comprising:

a decomposer configured to decompose an image into a set of hierarchical parts;

a hierarchy of classifiers, including multiple view-based classifiers, employed to detect features in the image and the hierarchical parts; and

at least one logical operation configured to logically combine a classification result from at least two of the classifiers to detect an occlusion state of the image by detennining that no face is detected in an image that has a detected person.

2. The system as recited in claim 1, wherein the occlusion state includes occlusion of a face.

3. The system as recited in claim 1, wherein the hierarchy of classifiers is applied to the image in an interleaved manner.

4. The system as recited in claim 3, wherein the image includes a sequence of images and each of a plurality of classifiers is applied to a next image of the sequence such that each classifier is applied to only one image at a time.

5. The system as recited in claim 4, wherein the classifiers are employed in an order for each successive image.

6. The system as recited in claim 1, wherein the at least one logical operation is configured to logically combine a classification result to permit a plurality of conditions to be met.

7. The system as recited in claim 1, further comprising a security feature configured to be responsive to the occlusion state.

8. The system as recited in claim 7, wherein the security feature includes a secured entry point and the occlusion state decides whether a person is authorized to enter the secured entry.

32. The occlusion detection system of claim 1, wherein the at least one logical operation includes at least one of AND, NOT, and OR.

9. A person detection system, comprising:

a face detector configured to detect a face in an input video sequence by interleaving of multiple view-based classifiers, the face detector outputting a face keyframe to be stored if a face is detected;

a person detector configured to detect a person in the input video sequence if the face detector fails to detect a face, the person detector outputting a person keyframe to be stored, if a person is detected in the input video sequence.

10. The system as recited in claim 9, wherein the face detector includes a hierarchy of classifiers employed in a rule-based combination to detect features in the image.

11. The system as recited in claim 9, wherein the person detector includes a hierarchy of classifiers employed in a rule-based combination to detect features in the image.

12. The system as recited in claim 9, further comprising a user interface configured to search based on keyframe selection that outputs face keyframes if the faces are not occluded, and person keyframes, if the person is present and the face is occluded.

13. The system as recited in claim 9, wherein the face detector is configured to track a face in a video input using correlation-based tracking.

14. The system as recited in claim 9, wherein the person detector is configured to track a person in a video input using correlation-based tracking.

15. The system as recited in claim 14, wherein the person detector is configured to detect a person in the video using interleaving of multiple view-based classifiers.

16. A method for occlusion detection, comprising:

decomposing an image into a set of hierarchical parts;

detecting features in the image and the hierarchical parts using a hierarchy of classifiers, including multiple view-based classifiers; and

logically combining classification results from at least two of the classifiers to detect an occlusion state of the image by determining that no face is detected in an image that has a detected person.

17. The method as recited in claim 16, wherein the occlusion state includes occlusion of a face.

18. The method as recited in claim 16, wherein the occlusion state includes a non-occluded face.

19. The method as recited in claim 16, further comprising interleaving usage of the hierarchy of classifiers applied to the image.

20. The method as recited in claim 19, wherein the image includes a sequence of images and each of a plurality of classifiers is applied to a next image of the sequence such that each classifier is applied to only one image at a time.

21. The method as recited in claim 20, wherein the classifiers are employed in an order for each successive image.

22. The method as recited in claim 16, wherein logically combining classification results permits a plurality of conditions to be met.

23. The method as recited in claim 16, further comprising providing access to an entry point responsive to the occlusion state.

24. The method as recited in claim 16, further comprising searching images based on keyframe selection that outputs face keyframes if the faces are not occluded, and person keyframes, if the person is present and the face is occluded.

25. The method as recited in claim 16, wherein detecting features includes tracking a face in a video input using correlation-based tracking.

26. The method as recited in claim 16, wherein detecting features includes interleaving of multiple view-based classifiers.

27. The method as recited in claim 16, wherein detecting features includes tracking a person in a video input using correlation-based tracking.

28. The method as recited in claim 16, further comprising detecting a face mask even if a face is fully occluded.

29. The method as recited in claim 16, further comprising generating real-time alerts for security in case a face mask is detected.

30. The method as recited in claim 16, further comprising generating real-time alerts for security in case a face mask is not detected.

31. A computer program product for occlusion detection comprising a non-transitory computer useable storage medium including a computer readable program, wherein the computer readable program when executed on a computer causes the computer to perform the steps of:

decomposing an image into a set of hierarchical parts;

detecting features in the image and the hierarchical parts using a hierarchy of classifiers, including multiple view-based classifiers; and

logically combining classification results from at least two of the classifiers to detect an occlusion state of the image by determining that no face is detected in an image that has a detected person.