Method and apparatus for classifying an object

• US 7,466,860 B2
• Filed: 08/26/2005
• Issued: 12/16/2008
• Est. Priority Date: 08/27/2004
• Status: Active Grant

• Current Assignee: ZAMA Innovations LLC
• Original Assignee: Sarnoff Corporation (SRI International, Inc.)
• Inventors: Chang, Peng, Camus, Theodore Armand
• Primary Examiner(s): Desire, Gregory M
• Application Number:
  US11/213,660
• Publication Number:
  US 20070237398A1
• Time in Patent Office: 1,208 Days
• Field of Search: 382/103, 382/154, 382/156, 382/159, 382/224
• US Class Current: 382/224
• CPC Class Codes: G06F 18/254 of classification results, e.g. of results related to same input data
• CPC Class Codes: G06V 20/58 Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
• CPC Class Codes: G06V 40/103 Static body considered as a whole, e.g. static pedestrian or occupant recognition

Abstract

A method and apparatus for classifying an object in an image is disclosed. A plurality of sub-classifiers is provided. An object is classified using input from each of the plurality of sub-classifiers.

Claims

1. A method of classifying an object in an image, comprising:

providing a plurality of sub-classifiers, the plurality of sub-classifiers comprising image-based classifiers and depth based classifiers, the image-based classifiers comprising one of a contour-based classifier and an edge-based classifier, the image-based classifiers identifying elements based on at least one image, the depth-based classifiers identifying elements based on depth relative to one of a sensor and at least one reference point, the contour-based classifiers identifying elements based on properties derived from significant intensity transitions of similar quality in an image along continuous long segments or curves, the edge-based classifiers identifying elements based on properties derived from significant intensity transitions of similar quality in an image along short, straight, continuous segments; and

classifying the object using input from each of the plurality of sub-classifiers.

2. The method of claim 1, wherein said classifying comprises combining the plurality of sub-classifiers using a Bayesian approach.

3. The method of claim 1, wherein said classifying comprises combining the plurality of sub-classifiers using a hybrid classifier.

4. The method of claim 3, wherein said hybrid classifier defines hypotheses;

and uses a likelihood ratio to decide which hypothesis to select.

5. The method of claim 4, wherein a classification decision is made according to:

H₀ if f_k>

0

H₁ otherwise

where H₀ and H₁, are the hypotheses and f_k is the likelihood ratio.

6. The method of claim 1, wherein the depth-based classifiers comprise measuring the peakness of a pedestrian model match.

7. The method of claim 6, wherein the peakness of a pedestrian model match is defined as the ratio of a first pedestrian 3D score and the sum of two pedestrian 3D scores that are adjacent to the first pedestrian 3D score.

8. The method of claim 1, wherein the depth-based classifiers comprise obtaining a ratio of a pedestrian 3D model score and a car 3D model score.

9. The method of claim 1, wherein the depth-based classifiers comprise an eccentricity test.

10. The method of claim 9, wherein the eccentricity test measures the flatness of a target in a depth map.

11. The method of claim 1, wherein the depth-based classifiers comprise one of a vehicle model three-dimensional block template matching score and a pedestrian model three-dimensional block template matching score.

12. An apparatus for classifying an object in an image, comprising:

means for providing a plurality of sub-classifiers, the plurality of sub-classifiers comprising image-based classifiers and depth based classifiers, the image-based classifiers comprising one of a contour-based classifier and an edge-based classifier, the image-based classifiers identifying elements based on at least one image, the depth-based classifiers identifying elements based on depth relative to one of a sensor and at least one reference point, the contour-based classifiers identifying elements based on properties derived from significant intensity transitions of similar quality in an image along continuous long segments or curves, the edge-based classifiers identifying elements based on properties derived from significant intensity transitions of similar quality in an image along short, straight, continuous segments; and

means for classifying the object using input from each of the plurality of sub-classifiers.

13. The apparatus of claim 12, wherein said classifying comprises combining the plurality of sub-classifiers using a Bayesian approach.

14. The apparatus of claim 12, wherein said classifying comprises combining the plurality of sub-classifiers using a hybrid classifier.

15. The apparatus of claim 14, wherein said hybrid classifier defines hypotheses;

and uses a likelihood ratio to decide which hypothesis to select.

16. The apparatus of claim 15, wherein a classification decision is made according to:

H₀ if f_k>

0

H₁ otherwise

where H₀ and H₁, are the hypotheses and f_k is the likelihood ratio.

17. The apparatus of claim 12, wherein the depth-based classifiers comprise one of a vehicle model three-dimensional block template matching score and a pedestrian model three-dimensional block template matching score.

18. The apparatus of claim 12, wherein the depth-based classifiers comprise measuring the peakness of a pedestrian model match.

19. The apparatus of claim 18, wherein the peakness of a pedestrian model match is defined as the ratio of a first pedestrian 3D score and the sum of two pedestrian 3D scores that are adjacent to the first pedestrian 3D score.

20. The apparatus of claim 12, wherein the depth-based classifiers comprise obtaining a ratio of a pedestrian 3D model score and a car 3D model score.

21. The apparatus of claim 12, wherein the depth-based classifiers comprise an eccentricity test.

22. The apparatus of claim 21, wherein the eccentricity test measures the flatness of a target in a depth map.

23. A computer-readable medium having stored thereon a plurality of instructions, the plurality of instructions including instructions which, when executed by a processor, cause the processor to perform the steps of a method for classifying an object in an image, comprising:

providing a plurality of sub-classifiers, the plurality of sub-classifiers comprising image-based classifiers and depth based classifiers, the image-based classifiers comprising one of a contour-based classifier and an edge-based classifier, the image-based classifiers identifying elements based on at least one image, the depth-based classifiers identifying elements based on depth relative to one of a sensor and at least one reference point, the contour-based classifiers identifying elements based on properties derived from significant intensity transitions of similar quality in an image along continuous long segments or curves, the edge-based classifiers identifying elements based on properties derived from significant intensity transitions of similar quality in an image along short, straight, continuous segments; and

classifying the object using input from each of the plurality of sub-classifiers.

24. The computer-readable medium of claim 23, wherein said classifying comprises combining the plurality of sub-classifiers using a hybrid classifier.

25. The computer-readable medium of claim 24, wherein said hybrid classifier defines hypotheses;

and uses a likelihood ratio to decide which hypothesis to select.

26. The computer-readable medium of claim 25, wherein a classification decision is made according to:

H₀ if f_k>

0

H₁ otherwise

where H₀ and H₁, are the hypotheses and f_k is the likelihood ratio.

27. The computer-readable medium of claim 23, wherein the depth-based classifiers comprise one of a vehicle model three-dimensional block template matching score and a pedestrian model three-dimensional block template matching score.

28. The computer-readable medium of claim 23, wherein said classifying comprises combining the plurality of sub-classifiers using a Bayesian approach.

29. The computer-readable medium of claim 23, wherein the depth-based classifiers comprise measuring the peakness of a pedestrian model match.

30. The computer-readable medium of claim 29, wherein the peakness of a pedestrian model match is defined as the ratio of a first pedestrian 3D score and the sum of two pedestrian 3D scores that are adjacent to the first pedestrian 3D score.

31. The computer-readable medium of claim 23, wherein the depth-based classifiers comprise obtaining a ratio of a pedestrian 3D model score and a car 3D model score.

32. The computer-readable medium of claim 23, wherein the depth-based classifiers comprise an eccentricity test.

33. The computer-readable medium of claim 32, wherein the eccentricity test measures the flatness of a target in a depth map.