Methods and apparatus for machine vision high accuracy searching

• US 5,602,937 A
• Filed: 06/01/1994
• Issued: 02/11/1997
• Est. Priority Date: 06/01/1994
• Status: Expired due to Term

• Current Assignee: Cognex Corporation
• Original Assignee: Cognex Corporation
• Inventors: Silver, William M., Bedrosian, David P.
• Primary Examiner(s): Mancoso, Joseph
• Assistant Examiner(s): DEL ROSSO, GERARD DOMNICK
• Application Number:
  US08/252,845
• Time in Patent Office: 986 Days
• Field of Search: 382/151, 382/144-148, 382/209
• US Class Current: 382/151
• CPC Class Codes: G06T 7/75 involving models

Abstract

An improved machine vision apparatus, of the type having a search element capable of estimating the location of a template image in a candidate image, includes an element that accepts signals representative of the template and of a candidate image. Those signals are applied to the search element to obtain its estimate of location. A correction element responds to the resulting approximation by the location-finder for generating a signal representative of more exacting location based on a known relationship between the location-finder output and more exacting locations of the template in a model of the candidate image.

Claims

1. In a machine vision apparatus of the type having

location finding means, responsive to an applied template image-representative signal and an applied candidate image-representative signal, for determining an approximate location, in the candidate image, of an object shown in the template image, and for generating an approximation signal representative of such location, the improvement for determining a more exact location in the candidate image of the object shown in the template image, comprising

A. means for inputting a signal representative of said template image and a signal representative of said candidate image,

B. runtime means for invoking said location finding means with said template image-representative signal and said candidate image-representative signal,

C. correction means, coupled to said location finding means, for generating a signal representative of a more exact location of said object in said candidate image as a function of an empirically-based relationship between an approximate location determined by said location finding means and a known location of said object, the empirically-based relationship determined from prior invocation of the location finding means with the template image-representative signal and one or more signals representative of a model candidate image at known locations.

2. In a machine vision apparatus according to claim 1, the further improvement wherein said correction means comprises

A. table means for storing values representative of an empirically-based relationship between approximation signals generated by said location finding means and corresponding more exact locations of said template, and

B. look-up means, coupled with said table means, for responding to a signal representative of a value represented by an approximation signal for generating a signal representative of a corresponding more exact value of said location.

3. In a machine vision apparatus according to claim 2, the further improvement wherein said table means includes means for storing values representative of more exact values of said location as indexed by values of approximation signals generated by said location finding means.

4. In a machine vision apparatus according to claim 3, the further improvement wherein said table means includes a first and second tables for storing, respectively, values representative of first and second coordinates of said more exact locations as indexed by values of approximation signals generated by said location finding means.

5. In a machine vision apparatus according to claim 1, the further improvement comprising training means, responsive to a signal representative of an object shown in said template image and a signal representative of a model candidate image, for generating signals representative of an empirically-based relationship between approximation signals generated by said location finding means and corresponding known locations of said template.

6. In a machine apparatus according to claim 5, the further improvement wherein said training means comprises means for invoking said location finding means with said template image-representative signal and with signals representative of successive model candidate image-representative signals for determining said empirically-based relationship.

7. In a machine apparatus according to claim 6, the further improvement wherein said training means comprises inversion means for responding to approximation signals generated by said location finding means to index values representative of locations corresponding to model candidate images by the values of approximation signals generated by said location finding means.

8. In a machine vision apparatus according to claim 6, the further improvement wherein said training means comprises image shifting means for shifting a signal representative of said model candidate image one or more times to generate respective ones of said successive model candidate image-representative signals.

9. In a machine vision apparatus according to claim 8, the further improvement wherein said image shifting means comprises means for shifting a signal representative of said model candidate image in increments corresponding to one or more fractions of a pixel to generate respective ones of said successive model candidate image-representative signals.

10. In a machine vision apparatus according to claim 6, the further improvement wherein said image shifting means comprises means for shifting by cubic convolution interpolation a signal representative of said model candidate image.

11. In a method for operating a machine vision apparatus of the type having

location finding means, responsive to an applied template image-representative signal and an applied candidate image-representative signal, for determining an approximate location of an object shown in said template image in said candidate image and for generating an approximation signal representative of such location,

the improvement for determining a more exact location, in the candidate image of an object shown in the template image, comprising

A. an inputting step for inputting a signal representative of said template image and a signal representative of said candidate image,

B. a runtime step for invoking said location finding means with said image-representative signal and said candidate image-representative signal,

C. correction step for generating a signal representative of a more exact location of said template image in said candidate image as a function of an empirically-based relationship between an approximate location determined by said location finding means and a known location of said object, the empirically-based relationship determined from prior invocation of the location finding means with the template image-representative signal and one or more signals representative of a model candidate image at known locations.

12. In a method according to claim 11, the further improvement wherein said correction step comprises

A. storing, in a table, values representative of an empirically-based relationship between approximation signals generated by said location finding means and corresponding known locations of said template, and

B. responding to a signal representative of a value represented by an approximation signal for looking up in said table a signal representative of a corresponding more exact value of said location.

13. In a method according to claim 12, the further improvement wherein said storing step includes the step of storing values representative of more exact locations as indexed by values of approximation signals generated by said location finding means.

14. In a method according to claim 13, the further improvement wherein said storing step comprises the step of storing in first and second tables, respectively, values representative of first and second coordinates of more exact values of said location as indexed by values of approximation signals generated by said location finding means.

15. In a method according to claim 11, the further improvement comprising a training step for responding to a signal representative of said template image and a signal representative of a model candidate image by generating signals representative of an empirically-based relationship between approximation signals generated by said location finding means and corresponding known locations of said template.

16. In a method according to claim 15, the further improvement wherein said training step comprises a step for invoking said location finding means with said template image-representative signal and with signals representative of successive model candidate image-representative signals for determining said empirically-based relationship.

17. In a method according to claim 16, the further improvement wherein said training step comprises an inversion step for responding to approximation signals generated by said location finding means to index values representative of locations corresponding to model candidate images by values of approximation signals generated by said location finding means.

18. In a method according to claim 16, the further improvement wherein said training step comprises an image shifting step for shifting a signal representative of said model candidate image a one or more times to generate respective ones of said successive model candidate image-representative signals.

19. In a method according to claim 18, the further improvement wherein said image shifting step comprises a step for shifting a signal representative of said model candidate image in increments corresponding to one or more fractions of a pixel to generate respective ones of said successive model candidate image-representative signals.

20. In a method according to claim 16, the further improvement wherein said image shifting step comprises a step for shifting by cubic convolution interpolation a signal representative of said model candidate image.