Method and system for aligning geometric object models with images

• US 6,804,416 B1
• Filed: 03/16/2001
• Issued: 10/12/2004
• Est. Priority Date: 03/16/2001
• Status: Expired due to Fees

• Current Assignee: Cognex Corporation
• Original Assignee: Cognex Corporation
• Inventors: Jacobson, Lowell, Negro, Jay, Segal, Leonid, Bachelder, Ivan
• Primary Examiner(s): PATEL, KANJIBHAI B
• Application Number:
  US09/809,026
• Time in Patent Office: 1,306 Days
• Field of Search: 382/151, 382/155, 382/159, 382/156, 382/294, 382/295, 382/305, 382/296, 382/312, 382/289, 382/190, 345/441, 345/649, 345/651, 345/420, 345/653, 345/654, 345/630
• US Class Current: 382/294
• CPC Class Codes: G06F 18/40 Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor

Abstract

A method and system is provided for aligning a geometric object model with a pixel image. A geometric object model representing an object is obtained first. Using the geometric object model of the object, a revised geometric object model is created. An alignment tool is then trained based on the revised geometric object model. The alignment model can be applied to a pixel image of the object so that the pose of the object can be computed by aligning the revised geometric object model with the image.

Claims

1. A method for aligning a geometric object model with a pixel image, said method comprising:

obtaining a geometric object model representing an object;

converting the geometric object model into a set of geometric elements;

selecting from the set of geometric elements a set of salient geometric elements by determining the acceptability of each of said geometric elements according to at least one criterion;

creating a revised geometric object model using the set of salient geometric elements;

training at least one alignment model based on the revised geometric object model;

acquiring a pixel image object representation; and

aligning said revised geometric object model with said pixel image object representation using said at least one alignment model.

2. The method of claim 1, wherein obtaining a geometric object model includes obtaining a CAD model, and wherein converting the geometric object model into a set of geometric elements includes removing parts of the CAD model.

3. The method of claim 1, wherein said converting the geometric object model into a set of geometric elements includes converting the geometric object model into a set of lines.

4. The method of claim 1, wherein said converting the geometric object model into a set of geometric elements includes converting the geometric object model into a set of arcs.

5. The method of claim 1, wherein said geometric elements comprise circular and elliptical arcs and lines.

6. The method of claim 1, wherein the at least one criterion includes:

visibility of a geometric element in said pixel image object representation.

7. The method of claim 1, wherein the the at least one criterion includes:

non-occlusion of a geometric element in said pixel image object representation.

8. The method of claim 1, wherein the at least one criterion includes:

detectability of geometric element in said pixel image object representation.

9. The method of claim 1, wherein the at least one criterion includes

whether a geometric elements is selected by a human operator using a graphical interface.

10. The method of claim 1, wherein said aligning includes:

extracting a set of image features from said pixel image representation of said object, each of said image features corresponding to a different salient geometric element included in said revised geometric object model;

matching said set of image features with corresponding salient geometric elements included in said revised geometric object model to produce a set of matched pairs between said set of image features and said corresponding salient geometric elements included in said revised geometric object model; and

estimating the pose of said object in said pixel image representation based on said set of matched pairs.

11. The method of claim 10, wherein matching said set of image features with corresponding salient geometric elements included in said revised geometric object model includes:

generating alignment performance information.

12. The method of claim 1, wherein aligning includes generating alignment performance information, and the method further includes:

using the alignment performance information and the revised object model to create a further revised geometric object model having a different set of geometric elements; and

using the further revised geometric object model to create a revised alignment model.

13. The method of claim 12, wherein using the alignment performance information to create a further revised geometric object model having a different set of geometric elements includes:

using the alignment performance information to remove geometric elements from the revised object model.

14. The method of claim 13, wherein using the alignment performance information to remove geometric elements from the revised object model includes:

removing geometric elements that are not adequately detected in an image during alignment.

15. A system for estimating the pose of an object in a pixel image, said system comprising:

a storage mechanism storing at least one geometric object model;

a converter to convert the at least one geometric object model to a set of geometric elements;

a selector to select a set of salient geometric elements from the set of geometric elements, the selector having a determiner to determine the acceptability of each of said geometric elements according to at least one criterion so as to provide the set of salient geometric elements;

a revised geometric object model builder to build a revised geometric object model using the set of salient geometric elements;

a training unit to train at least one alignment tool based on the revised geometric object model;

an image acquisition mechanism to obtain a pixel image object representation; and

an alignment system to align said revised geometric object model with said pixel image object representation using said at least one alignment tool.

16. The system of claim 15, wherein said geometric object model includes a CAD model, and wherein the converter converts the geometric object model into a set of geometric elements be removing parts of the CAD model.

17. The system of claim 15, wherein said at least one criterion includes the visibility of said geometric elements in said pixel image representation.

18. The system of claim 15, wherein said at least one criterion includes the non-occlusiveness of said geometric elements in said pixel image representation.

19. The system of claim 15, wherein said at least one criterion includes the detectability of said geometric elements in said pixel image representation.

20. The system of claim 15, wherein said at least one criterion includes whether said geometric elements are selected by a human operator through a graphical editor.

21. The system of claim 15, wherein said alignment system includes:

an extractor to extract a set of image features from said pixel image representation of said object, each of said image features corresponding to a different salient geometric element included in said revised geometric object model;

a matcher to match said set of image features with corresponding salient geometric elements included in said revised geometric object model to produce a set of matched pairs between said set of image features and said corresponding salient geometric elements included in said revised geometric object model; and

an estimator to estimate the pose of said object in said pixel image representation based on said set of matched pairs.

22. The system of claim 21, wherein the matcher provides feedback information to the selector regarding geometric elements in the revised geometric object model that have not adequately matched to said set of image features.

23. The system of claim 15, wherein the alignment system provides feedback information to the selector regarding geometric elements in the revised geometric object model that have not adequately matched to said set of image features.

24. A medium having information recorded thereon, such that when said information is read and executed by a computer, the computer is caused to:

obtain a geometric object model representing an object;

convert the geometric object model into a set of geometric elements;

select from the set of geometric elements a set of salient geometric elements by determining the acceptability of each of said geometric elements according to at least one criterion;

create a revised geometric object model using the set of salient geometric elements;

train at least one alignment model based on the revised geometric object model;

acquire a pixel image object representation; and

align said revised geometric object model with said pixel image object representation using said at least one alignment model.

25. The medium of claim 24, wherein said information recorded on said medium further causing said computer to:

extract a set of image features from said pixel image representation of said object, each of said image features corresponding to a different salient geometric element included in said revised geometric object model;

match said set of image features with corresponding salient geometric elements included in said revised geometric object model to produce a set of matched pairs between said set of image features and said corresponding salient geometric elements included in said revised geometric object model; and

estimate the pose of said object in said pixel image representation based on said set of matched pairs.