Method for fast, robust, multi-dimensional pattern recognition

• US 8,265,395 B1
• Filed: 12/21/2004
• Issued: 09/11/2012
• Est. Priority Date: 07/13/1998
• Status: Expired due to Term

• Current Assignee: Cognex Technology & Investment Corporation (Cognex Corporation)
• Original Assignee: Cognex Corporation
• Inventors: Silver, William M., McGarry, E. John, Hill, Matt, Foster, Nigel J., Foster, Willard
• Primary Examiner(s): Koziol, Stephen R
• Application Number:
  US11/023,230
• Time in Patent Office: 2,821 Days
• Field of Search: 382/118, 382/141, 382/149, 382/181, 382/190
• US Class Current: 382/181
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06V 10/469 Contour-based spatial representations, e.g. vector-coding
• CPC Class Codes: G06V 10/752 Contour matching

Abstract

Disclosed is a method for determining the absence or presence of one or more instances of a predetermined pattern in an image, and for determining the location of each found instance within a multidimensional space. A model represents the pattern to be found, the model including a plurality of probes. Each probe represents a relative position at which a test is performed in an image at a given pose, each such test contributing evidence that the pattern exists at the pose. The method further includes a comparison of the model with a run-time image at each of a plurality of poses. A match score is computed at each pose to provide a match score surface. Then, the match score is compared with an accept threshold, and used to provide the location any instances of the pattern in the image.

Claims

1. A method for refining search results representing possible instances of a pattern in an image having a square grid of pixels and associated positions, the search results being characterized by at least one non-translation degree of freedom (DOF), the method comprising the steps of:

using a data processing device to run a program to, for each DOF instance in a DOF list, divide a corresponding DOF step size by 2, each DOF instance including the corresponding DOF step size and a 2D coordinate transform function based on a DOF associated with the DOF instance; and

providing a refined search result, using the data processing device, for each search result to be refined in a result list, by performing a fine-scan procedure using the DOF list, 2D coordinate transform functions included in DOF instances in the DOF list, and the search result to be refined.

2. The method of claim 1, wherein the method is repeated by the data processing device to provide further refinement.

3. The method of claim 1, wherein the DOF list is a list of generalized DOF instances.

4. The method of claim 1, wherein each generalized DOF instance includes:

a parameter range including a low limit value and a high limit value that specifies a range of allowed parameter values to be searched,

wherein the corresponding DOF step size specifies the maximum allowable separation between parameter values used in scanning the generalized DOF, and the corresponding 2D coordinate transform function maps a real-value parameter to a corresponding 2D coordinate transform value.

5. The method of claim 1, wherein each search result includes a position and a match score.

6. The method of claim 1, wherein the corresponding 2D coordinate transform is a function of a real argument.

7. The method of claim 1, wherein the corresponding 2D coordinate transform is a matrix.

8. The method of claim 1, wherein, when the data processing device is performing the fine-scan procedure, the data processing device analyzes and updates all of the translation and non-translation degrees of freedom of the pose of the search result so as to provide a refined search result.

9. The method of claim 1, wherein the DOF list starts with a current generalized DOF instance.

10. The method of claim 1, wherein the corresponding 2D coordinate transform constructs a partial pose.

11. The method of claim 1, wherein the data processing device is programmed to perform the fine-scan procedure by performing a fine-scan in x-y if the DOF list is empty.

12. The method of claim 1, wherein the fine-scan procedure is a recursive function.

13. The method of claim 3, wherein the data processing device is programmed to perform the fine-scan procedure by operating on an outermost generalized DOF instance which is the first element of the DOF list, and calls itself to operate on elements of the DOF list after the first element.

14. The method of claim 12, wherein at each outer level of recursion, a partial pose is constructed.

15. The method of claim 14, wherein the data processing device is programmed to construct the partial pose by composing a current mapper transform with an input map constructed by recursion levels representing outer generalized DOFs, and passing the partial pose along to recursion levels representing inner generalized DOFs.

16. The method of claim 13, wherein if a current generalized DOF is determined to be non-fixed, the data processing device is programmed to perform a hill climbing process.

17. The method of claim 16, wherein the data processing device is programmed to perform the hill climbing process by performing the steps of:

evaluating poses at plus one step size and minus one step size from a current generalized DOF parameter value; and

testing whether a score improves upon adding one step size or upon subtracting one step size from the current generalized DOF parameter value.

18. The method of claim 17, wherein the data processing device is programmed to further perform the step:

if the score does not improve, performing an interpolation.

19. The method of claim 17, wherein the data processing device is programmed to further perform the step of:

if the score improves upon adding one step size to the current generalized DOF parameter value, testing whether adding one step size exceeds a high limit of a search range.

20. The method of claim 19, wherein the data processing device is programmed to further perform the step of:

if adding one step size does not exceed a high limit of a search range, again adding one step size to the current generalized DOF parameter value; and

evaluating a new set of poses.

21. The method of claim 20, wherein the data processing device is programmed to further perform the step of:

if the score improves upon adding one step size to the current generalized DOF parameter value, again adding one step size to the current generalized DOF parameter value;

else

scanning terminates.

22. The method of claim 11, wherein the data processing device is programmed to perform the step of performing a fine-scan in x-y by evaluating a match function at each pixel position within an approximately circular region of the square grid, the approximately circular region being centered on a translation of the result being refined.

23. The method of claim 22, wherein the data processing device is programmed to further perform the step of:

determining a highest value of the match function at a pixel position; and

interpolating the pixel position with a three-point parabolic interpolation using the north-south neighbors of the pixel position, and the east-west neighbors of the pixel position, so as to provide a refined result at the interpolated pixel position having the highest value.

24. The method of claim 22, wherein the approximately circular region includes either 12, 32, or 52 pixel positions.

25. A method for refining search results representing possible instances of a pattern in an image having a square grid of pixels and associated positions, the search results being characterized by at least one non-translation degree of freedom (DOF), the method comprising the steps of:

using a data processing device to run a program to, for each DOF instance in a DOF list, divide a corresponding DOF step size by a value greater than one thereby providing a revised DOF step size, each DOF instance including the corresponding DOF step size and a 2D coordinate transform function based on a DOF associated with the DOF instance; and

providing a refined search result, using the data processing device to, for each search result to be refined in a result list, by performing a fine-scan procedure using the DOF list, 2D coordinate transform functions included in DOF instances in the DOF list, modified DOF step sizes included in DOF instances in the DOF list, and the search result to be refined.

26. In a two-dimensional pattern matching data processing product for refining search results representing possible instance of a pattern in an image, the search results being characterized by at least one non-translation degree of freedom (DOF) the product tangibly embodied in a non-transitory and data processing device readable medium, the product comprising instructions being operable to cause a data processing apparatus to:

divide, for each DOF instance in a DOF list, a corresponding DOF step size by a value greater than one there by providing a revised DOF step size, each DOF instance including the corresponding DOF step size and a 2D coordinate transform function based on a DOF associated with the DOF instance; and

provide a refined search result for each search result in a result list, by performing a fine-scan procedure using the DOF list, 2D coordinate transform functions included in DOF instances in the DOF list, modified DOF step sizes included in DOF instances in the DOF list, and the search result to be refined.

27. The product of claim 26, wherein the fine-scan procedure operates on an outermost DOF instance which is the first element of the DOF list, and calls itself to operate on elements of the DOF list after the first element.

28. The product of claim 27, wherein if a current DOF is determined to be non-fixed, the fine scan procedure further comprises:

evaluating poses at plus one step size and minus one step size from a current DOF parameter value; and

testing whether a score improves upon adding one step size or upon subtracting one step size from the current DOF parameter value.

29. The product of claim 26, wherein the fine-scan procedure further comprises:

recursively performing the fine-scan procedure, at each outer level of recursion, constructing a partial pose.

30. The product of claim 29, wherein constructing the partial pose further comprises:

composing a current mapper transform with an input map constructed by recursion levels representing outer DOFs, and passing the partial pose along to recursion levels representing inner DOFs.

31. The product of claim 26, wherein each DOF instance comprises:

a parameter range including a low limit value and a high limit value that specifies a range of allowed parameter values to be searched; and

wherein the corresponding DOF step size specifies the maximum allowable separation between parameter values used in scanning DOF, and the corresponding 2D coordinate transform function maps a real-valued parameter to a corresponding 2D coordinate.