Method and apparatus for recognition of symbols in images of three-dimensional scenes

• US 7,738,706 B2
• Filed: 12/30/2004
• Issued: 06/15/2010
• Est. Priority Date: 09/22/2000
• Status: Active Grant

• Current Assignee: SRI International, Inc.
• Original Assignee: SRI International, Inc.
• Inventors: Aradhye, Hrishikesh, Myers, Gregory K.
• Primary Examiner(s): Alavi, Amir
• Application Number:
  US11/027,846
• Publication Number:
  US 20050180632A1
• Time in Patent Office: 1,993 Days
• Field of Search: 382/105, 382/143, 382/144, 382/147, 382/154, 382/173, 382/174, 382/176, 382/181, 382/202, 382/229, 382/290, 382/295, 382/296, 382/254, 382260-264, 382274-275
• US Class Current: 382/182
• CPC Class Codes: G06V 20/63 Scene text, e.g. street names
• CPC Class Codes: G06V 30/1478 of characters or characters lines

Abstract

An apparatus and a concomitant method for rectification and recognition of symbols (e.g., to correct for the effects of perspective distortion, rotation and/or scale) in images of three-dimensional scenes is provided. The present method locates a reference region lying in a common plane with a symbol to be recognized, where the reference region represents an image of a planar object having assumed (e.g., known or standard) geometry and dimensions, along with at least four easily detectable correspondence points within that geometry. An image of the common plane is then rectified in three dimensions in accordance with the assumed dimensions of the reference region in order to produce a transformed (e.g., rectified) image of the symbol, which is significantly easier to recognize than the original imaged appearance of the symbol using existing recognition methods.

Claims

1. Method for rectifying at least one target object in a captured imagery, said method comprising:

using a processor to perform steps comprising;

locating a reference region lying in a common plane with an image of said at least one target object, said reference region representing an image of a reference object that is defined by a bounding shape having geometry and dimensions that are known to said processor, where said image of said at least one target object lies outside of said bounding shape; and

rectifying an image of said common plane in accordance with said geometry and dimensions of said bounding shape.

2. The method of claim 1, wherein said rectifying is performed in three dimensions.

3. The method of claim 2, wherein said rectifying comprises:

applying a projective transformation to said image of said common plane.

4. The method of claim 1, further comprising:

recognizing said at least one target object in said rectified image of said common plane.

5. The method of claim 4, wherein said recognizing comprises:

applying image recognition processing to said rectified image to recognize said at least one target object in the captured imagery.

6. The method of claim 1, wherein said reference object is substantially quadrilateral in shape.

7. The method of claim 6, wherein said reference object is an automobile license plate.

8. The method of claim 6, wherein said reference object is a business card.

9. The method of claim 6, wherein said reference object is an identification card.

10. The method of claim 1, wherein said common plane is orientated at an oblique angle relative to an optical axis of a device used to capture said captured imagery.

11. A computer readable medium containing an executable program for rectifying at least one target object in a captured imagery, where the program performs steps of:

locating a reference region lying in a common plane with an image of said at least one target object, said reference region representing an image of a reference object that is defined by a bounding shape having geometry and dimensions that are known to a processor, where said image of said at least one target object lies outside of said bounding shape; and

rectifying an image of said common plane in accordance with said geometry and dimensions of said bounding shape.

12. The computer readable medium of claim 11, wherein said rectifying is performed in three dimensions.

13. The computer readable medium of claim 12, wherein said rectifying comprises:

applying a projective transformation to said image of said common plane.

14. The computer readable medium of claim 11, further comprising:

recognizing said at least one target object in said rectified image of said common plane.

15. The computer readable medium of claim 14, wherein said recognizing comprises:

applying image recognition processing to said rectified image to recognize said at least one target object in the captured imagery.

16. The computer readable medium of claim 11, wherein said reference object is substantially quadrilateral in shape.

17. The computer readable medium of claim 16, wherein said reference object is an automobile license plate.

18. The computer readable medium of claim 16, wherein said reference object is a business card.

19. The computer readable medium of claim 16, wherein said reference object is an identification card.

20. Apparatus for rectifying at least one target object in a captured imagery, said apparatus comprising:

means for locating a reference region lying in a common plane with an image of said at least one target object, said reference region representing an image of a reference object that is defined by a bounding shape having geometry and dimensions that are known to said apparatus, where said image of said at least one target object lies outside of said bounding shape; and

means for rectifying an image of said common plane in accordance with said geometry and dimensions of said bounding shape.

21. The apparatus of claim 20, wherein said means for rectifying rectifies said image of said common plane in three dimensions.

22. Method for rectifying at least one target object in a captured imagery, said method comprising:

using a processor to perform steps comprising;

locating a reference region lying in a common plane with an image of said at least one target object, said reference region representing an image of a reference object that is distinct from said image of said at least one target object and has geometry and dimensions that are known to the processor; and

rectifying along three axes an image of said common plane in accordance with said geometry and dimensions.

23. The method of claim 22, wherein said image of said reference object is defined by a bounding shape, and said image of said at least one target object lies outside of said bounding shape.

24. The method of claim 22, wherein said common plane is orientated at an oblique angle relative to an optical axis of a device used to capture said captured imagery.