Determining the Uniqueness of a Model for Machine Vision

• US 20120170834A1
• Filed: 12/29/2010
• Published: 07/05/2012
• Est. Priority Date: 12/29/2010
• Status: Active Grant

• Current Assignee: Cognex Corporation
• Original Assignee: Cognex Corporation
• Inventors: Wang, Xiaoguang, Jacobson, Lowell
• Granted Patent: US 8,542,905 B2
• Time in Patent Office: Days
• US Class Current: 382/159
• CPC Class Codes: G06F 18/2193 based on specific statistical tests
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries

Abstract

Described are methods and apparatuses, including computer program products, for determining model uniqueness with a quality metric of a model of an object in a machine vision application. Determining uniqueness involves receiving a training image, generating a model of an object based on the training image, generating a modified training image based on the training image, determining a set of poses that represent possible instances of the model in the modified training image, and computing a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image.

Claims

1. A computerized method for determining a quality metric of a model of an object in a machine vision application, the method comprising:

receiving a training image;

generating a model of an object based on the training image;

generating a modified training image based on the training image;

determining a set of poses that represent possible instances of the model in the modified training image; and

computing a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image.

2. The method of claim 1 further comprising computing at least a primary score and a secondary score for at least a portion of the set of poses.

3. The method of claim 2 wherein computing the quality metric is based on the primary score and the secondary score.

4. The method of claim 2 wherein computing the quality metric is based on a distribution of the computed scores for the portion of the set of poses.

5. The method of claim 1 wherein the model comprises a geometric description of the object in the training image.

6. The method of claim 1 wherein the model comprises a portion of the training image.

7. The method of claim 1 wherein generating the modified training image comprises adding noise to the training image.

8. The method of claim 7 wherein the noise comprises amplifier noise, salt-and-pepper noise, shot noise, quantization noise, film grain noise, non-isotropic noise, or any combination thereof.

9. The method of claim 7 wherein the noise is added to one or more pixels in the training image.

10. The method of claim 1 wherein generating the modified training image comprises transforming the training image by one or more degrees-of-freedom of rotation, translation, scale, skew, aspect ratio, or any combination thereof.

11. The method of claim 1 wherein generating the modified training image comprises changing the resolution of the training image.

12. The method of claim 1 further comprising:

generating a plurality of modified training images based on the training image; and

determining a set of poses for each of the plurality of modified training images, each set of poses representing possible instances of the model in one of the plurality of modified training images;

wherein computing the quality metric is further based on an evaluation of the sets of poses, determined from the plurality of modified training images, with respect to the modified training image.

13. The method of claim 12 further comprising computing at least primary scores and secondary scores for at least a portion of the set of poses and at least portions of each of the sets of poses computed from the plurality of modified training images.

14. The method of claim 13 wherein computing the quality metric of the model is based on the secondary scores.

15. The method of claim 13 wherein computing the quality metric of the model is based on a distribution of the computed scores for the portion of the set of poses and distributions of the computed scores for the portions of the sets of poses computed from the plurality of modified training images.

16. The method of claim 1 further comprising modifying a baseline model parameter, wherein generating the model is based on the modified baseline model parameter.

17. The method of claim 16 wherein the baseline model parameter comprises an elasticity parameter, a grain limit or granularity, a coarse-value acceptance fraction, a contrast threshold, an edge-value threshold, trainClientFromPattern or any combination thereof.

18. A computer program product, tangibly embodied in a machine-readable storage device, the computer program product including instructions being operable to cause a data processing apparatus to:

receive a training image;

generate a model of an object based on the training image;

generate a modified training image based on the training image;

determine a set of poses that represent possible instances of the model in the modified training image; and

compute a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image.

19. A system for determining a quality metric of a model of an object in a machine vision application, the system comprising:

interface means for receiving a training image;

model generating means for generating a model of an object based on the training image;

image modifying means for generating a modified training image based on the training image;

processor means for determining a set of poses that represent possible instances of the model in the modified training image; and

processor means for computing a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image.

20. A machine vision system for determining a quality metric of a model of an object in a machine vision application, the system comprising:

an interface for receiving a training image;

a model generating module for generating a model of an object based on the training image;

an image processing module for generating a modified training image based on the training image;

a run-time module for determining a set of poses that represent possible instances of the model in the modified training image; and

a quality-metric module for computing a quality metric of the model based on an evaluation of the set of poses with respect to the modified training image.