System and method for determining the position of an object in three dimensions using a machine vision system with two cameras

• US 6,728,582 B1
• Filed: 12/15/2000
• Issued: 04/27/2004
• Est. Priority Date: 12/15/2000
• Status: Active Grant

• Current Assignee: Cognex Corporation
• Original Assignee: Cognex Corporation
• Inventors: Wallack, Aaron
• Primary Examiner(s): Patel, Ramesh
• Assistant Examiner(s): PHAM, THOMAS K
• Application Number:
  US09/738,957
• Time in Patent Office: 1,229 Days
• Field of Search: 700/47, 700/56, 700/58, 700/61, 700/64, 700/66, 700/118, 700/163, 700/186-187, 382/154, 382/153, 382/159, 382/181
• US Class Current: 700/64
• CPC Class Codes: G05B 19/41875 characterised by quality surveillance of production
• CPC Class Codes: G05B 2219/37567 3-D vision, stereo vision, with two cameras
• CPC Class Codes: G05B 2219/50064 Camera inspects workpiece for errors, correction of workpiece at desired position
• CPC Class Codes: Y02P 90/02 Total factory control, e.g. smart factories, flexible manufacturing systems [FMS] or integrated manufacturing systems [IMS]

Abstract

A system and method for estimating the position of an object in three dimensions uses two cameras interconnected with a machine vision search tool. The search tool is capable of registering transformation of a pattern in at least two translational degrees of freedom, along the image plane, and at least one non-translational degree of freedom along a camera axis perpendicular to the image plane. The tool can be a rotation/scale-invariant (RSIS) search tool. A nominal position for each camera'"'"'s acquired image of the object is determined and a set of uncertainty vectors along each of the degrees of freedom is generated. The vectors are weighted, with the respective vectors along the camera axis being weighted more-highly than orthonormal vectors within the image plane. The weighted vectors are combined to define an error. The defined error is compared according to a least squares technique with selected three-dimensional position estimates and the minimum sum error is used to identify the best three-dimensional position estimate.

Claims

1. A system for estimating a three-dimensional position of an object using a machine vision system that acquires an image of the object containing a pattern that matches a trained pattern, the system comprising:

a search tool adapted to register transformation of a pattern by at least two translational degrees of freedom and at least one non-translational degree of freedom, the search tool providing, from the image based upon the trained pattern, at least a first search tool result relative to a first axis of viewing the object and a second search tool result relative to a second axis of viewing the object;

a combiner that combines the search tool results so as to determine an estimated three dimensional position of the object, the combiner being constructed and arranged to determine a first nominal coordinate position for the object along each of the at least two translational degrees of freedom and the at least one non-translational degree of freedom and a second nominal coordinate position for the object along each of the at least two translational degrees of freedom and the at least one non-translational degree of freedom, and to derive the estimated three-dimensional position of the pattern based upon combining predicted uncertainty in the first nominal coordinate position and the second nominal coordinate position by computing (a) a respective first direction unit vector along the first axis and a second direction unit vector along the second axis, each being respectively, representative of a predetermined uncertainty in the first nominal coordinate position along the first axis and the second nominal coordinate position along the second axis, and (b) a first pair of orthonornal vectors and (c) a second pair of orthonormal vectors lying within the respective image plane, each being representative of the predetermined uncertainty; and

the combiner being further constructed and arranged to perform an estimation based upon matching a combination of weighted values for all of the first direction unit vector, the second direction unit vector, the first pair of orthonormal vectors and the second pair of orthonormal vectors with a selected three-dimensional position estimate so as to compute an optimal three-dimensional position estimate.

2. The system as set forth in claim 1 wherein each of the first search tool result and the second search tool result are generated from views of the object using at least a first camera and a second camera spatially separated from the first camera, respectively.

3. The system as set forth in claim 1 wherein the combiner is constructed and arranged to perform the estimation by a least squares summing so as to compute the optimal estimate based upon a minimum sum error between the combination of weighted values and the selected three-dimensional position estimate.

4. The system as set forth in claim 1 wherein the search tool comprises a search tool that registers transformation of a trained pattern by at least two translational degrees of freedom and at least two non-translational degrees of freedom.

5. A method for estimating a three-dimensional position of an object using a machine vision system that acquires an image of the object containing a pattern that matches a trained pattern, the method comprising:

providing, with a search tool adapted to register transformation of a pattern by at least two translational degrees of freedom and at least one non-translational degree of freedom, from the image based upon the trained pattern, at least a first search tool result relative to a first axis of viewing the object and a second search tool result relative to a second axis of viewing the object;

combining the search tool results so as to determine an estimated three dimensional position of the object, including determining a first nominal coordinate position for the object alone each of the at least two translational degrees of freedom and the at least one non-translational degree of freedom and a second nominal coordinate position for the object along each of the at least two translational degrees of freedom and the at least one non-translational degree of freedom, and deriving the estimated three-dimensional position of the pattern based upon combining predicted uncertainty in the first nominal coordinate position and the second nominal coordinate position by computing (a) a respective first direction unit vector along the first axis and a second direction unit vector along the second axis, each being respectively, representative of a predetermined uncertainty in the first nominal coordinate position along the first axis and the second nominal coordinate position along the second axis, and (b) a first pair of orthonormal vectors and (c) a second pair of orthonormal vectors lying within the respective image plane, each being representative of the predetermined uncertainty; and

performing an estimation based upon matching a combination of weighted values for all of the first direction unit vector, the second direction unit vector, the first pair of orthonormal vectors and the second pair of orthonormal vectors with a selected three-dimensional position estimate so as to compute an optimal three-dimensional position estimate.

6. The method as set forth in claim 5 further comprising generating each of the first search tool result and the second search tool result using at least a first camera and a second camera spatially separated from the first camera, respectively.

7. The method as set forth in claim 5 wherein the step of performing the estimation includes performing a least squares summing so as to compute the optimal estimate based upon a minimum sum error between the combination of weighted values and the selected three-dimensional position estimate.

8. The method as set forth in claim 5 wherein the search tool registers transformation of a trained pattern by at least two translational degrees of freedom and at least two non-translational degrees of freedom.

9. The method as set forth in claim 8 wherein each of the at least two search tool results are derived from each of at least two respective images generated by at least two respective spatially separated cameras of a solitary instance of the trained pattern on the object.

10. A computer-readable medium including program instructions executed on a computer for estimating a three-dimensional position of an object using a machine vision system that acquires an image of the object containing a pattern that matches a trained pattern, the computer-readable medium comprising program instructions for performing the steps of:

providing, with a search tool adapted to register transformation of a pattern by at least two translational degrees of freedom and at least one non-translational degree of freedom, from the image based upon the trained pattern at least a first search tool result relative to a first axis of viewing the object and a second search tool result relative to a second axis of viewing the object;

combining the search tool results so as to determine an estimated three dimensional position of the object, including determining a first nominal coordinate position for the object along each of the at least two translational degrees of freedom and the at least one non-translational degree of freedom and a second nominal coordinate position for the object along each of the at least two translational degrees of freedom and the at least one non-translational degree of freedom, and deriving the estimated three-dimensional position of the pattern based upon combining predicted uncertainty in the first nominal coordinate position and the second nominal coordinate position by computing (a) a respective first direction unit vector along the first axis and a second direction unit vector along the second axis, each being respectively, representative of a predetermined uncertainty in the first nominal coordinate position along the first axis and the second nominal coordinate position along the second axis, and (b) a first pair of orthonormal vectors and (c) a second pair of orthonornal vectors lying within the respective image plane, each being representative of the predetermined uncertainty; and

performing an estimation based upon matching a combination of weighted values for all of the first direction unit vector, the second direction unit vector, the first pair of orthonormal vectors and the second pair of orthonormal vectors with a selected three-dimensional position estimate so as to compute an optimal three-dimensional position estimate.

11. The computer-readable medium as set forth in claim 10 further comprising generating each of the first search tool result and the second search tool result using at least a first camera and a second camera spatially separated from the first camera, respectively.

12. The computer-readable medium as set forth in claim 10 wherein the step of performing the estimation includes performing a least squares summing so as to compute the optimal estimate based upon a minimum sum error between the combination of weighted values and the selected three-dimensional position estimate.

13. The computer-readable medium as set forth in claim 10 wherein the search tool registers transformation of a trained pattern by at least two translational degrees of freedom and at least two non-translational degrees of freedom.

14. The computer-readable medium as set forth in claim 10 wherein each of the first search tool result and the second search tool result are derived from each of at least two respective images generated by at least two respective spatially separated cameras of a solitary instance of the trained pattern on the object.