Single step coarse registration and inspection of circular objects

• US 5,818,443 A
• Filed: 05/06/1996
• Issued: 10/06/1998
• Est. Priority Date: 05/06/1996
• Status: Expired due to Term

• Current Assignee: Cognex Corporation
• Original Assignee: Cognex Corporation
• Inventors: Schott, Jean-Pierre
• Primary Examiner(s): Couso, Jose L.
• Application Number:
  US08/643,174
• Time in Patent Office: 883 Days
• Field of Search: 382/169, 382/170, 382/171, 382/172, 382/151, 382/159, 382/209, 382/220, 382/221, 382/289, 382/294, 382/293, 382/144, 382/145
• US Class Current: 382/141
• CPC Class Codes: G06T 2207/30164 Workpiece; Machine component
• CPC Class Codes: G06T 7/001 using an image reference approach
• CPC Class Codes: G06T 7/30 Determination of transform parameters for the alignment of images, i.e. image registration

Abstract

A very fast, memory efficient, single-step solution for coarse alignment with determination of angular displacement, and inspection for surface defects on circular objects with non-symmetric patterns and random orientation is described. A suitable geometric partition model of an image is constructed that includes a plurality of spatial bins that can be used for rotation estimation and defect detection. In one embodiment for inspection of circular objects, the geometric partitioning includes 256 spatial bins constituted by concentric circular rings divided into sectors. The number of sectors and thickness of the rings are such that the approximate number of pixels in each region is the same. A set of numbers that encode rotational position and gray level information on the plurality of spatial bins of a reference object are acquired during training and used during run-time wherein the same information relative to a circular object being inspected is computed, and the data are correlated to determine both the coarse rotation and gray level difference in pertinent ones of the plurality of spatial bins.

Claims

1. A method of inspecting a circular object to effect angular displacement determination and defect detection, using a machine vision system capable of storing a train-time image of a standard sample of said circular object and capturing a run-time image of said circular object to compare against train-time pixel information, said method comprising the steps of:

providing a geometric partition model of said circular object, said geometric partition model comprising spatial bins, each of said spatial bins comprising a plurality of pixels, each of said plurality of pixels being assigned one of a plurality of said labels corresponding to a respective spatial bin;

generating a train-time spatial histogram of said train-time image in accordance with said geometric partition model wherein each pixel in said train-time image is assigned a train-time spatial histogram label selected from one of said plurality of labels, said train-time spatial histogram comprising a plurality of train-time sums of grey level values of pixels of said train-time image in respective ones of said spatial bins;

generating a run-time spatial histogram of said run-time image in accordance with said geometric partition model wherein each pixel in said run-time image is assigned a run-time spatial histogram label selected from one of said plurality of labels designating said respective one of said spatial bins, said run-time spatial histogram comprising a plurality of run-time sums of grey level values of pixels of said run-time image in respective ones of said spatial bins;

determining an estimated angular displacement of said run-time image relative to said train-time image to provide an aligned run-time spatial histogram in accordance with said geometric partition model, that is aligned with said train-time spatial histogram in accordance with said estimated angular displacement; and

comparing said train-time spatial histogram and said aligned run-time spatial histogram to determine a difference therebetween to provide an estimate of defects in said circular object.

2. The method of claim 1 wherein said step of determining said estimated angular displacement is performed by computing a plurality of sums of absolute differences between said train-time spatial histogram and said run-time spatial histogram.

3. The method of claim 1 wherein said step of determining an angular displacement further includes the step of analyzing selected ones of said plurality of spatial bins to determine respective angular offsets of said selected ones of said plurality of spatial bins to derive said estimated angular displacement of said run-time image in accordance with at least some of said respective angular offsets.

4. The method of claim 1 further comprising a coarse defect detection operation comprising the steps of:

generating a train-time intensity histogram of said train-time image in accordance with a don'"'"'t care mask, wherein said train-time intensity histogram comprises a plurality of train-time intensity bins each identified by a respective train-time bin address that corresponds to a respective run-time grey level value, and each of said plurality of train-time intensity bins includes a train-time bin value that is equal to a number of pixels of said train-time image that have a train-time grey level value corresponding to said respective train-time bin address;

generating a run-time intensity histogram of said run-time image in accordance with said don'"'"'t care mask, wherein said run-time intensity histogram comprises a plurality of run-time intensity bins each identified by a respective run-time bin address that corresponds to a respective run-time grey level value, and each of said plurality of run-time intensity bins includes a run-time bin value that is equal to a number of pixels of said run-time image that have a run-time grey level value corresponding to said respective run-time bin address;

computing a sum of absolute differences between said traintime intensity histogram and said run-time intensity histogram;

selecting a threshold of acceptability of said sum of absolute differences between said train-time intensity histogram and said run-time intensity histogram; and

classifying said circular object as a function of said threshold of acceptability and said sum of absolute differences between said train-time intensity histogram and said run-time intensity histogram.

5. The method of claim 1 further comprising a coarse defect detection comprising the steps of:

after said steps of generating said run-time spatial histogram and said train-time spatial histogram,

computing a train-time summation histogram by summing selected ones of said train-time spatial bins;

computing a run-time summation histogram by summing selected ones of said run-time spatial bins;

computing a sum of absolute differences between said train-time summation histogram and said run-time summation histogram;

selecting a threshold of acceptability of said sum of absolute differences between said train-time summation histogram and said run-time summation histogram; and

classifying said circular object as a function of said threshold of acceptability and said sum of absolute differences between said train-time summation histogram and said run-time summation histogram.

6. The method of claim 1 wherein said geometric partition model comprises a plurality of concentric rings, and each of said plurality of concentric rings is divided into a plurality of sectors, each of said sectors corresponding to a respective one of said plurality of spatial bins.

7. The method of claim 1 wherein each of said plurality of spatial bins comprises approximately an equal number of pixels.

8. The method of claim 4 wherein said don'"'"'t care mask of said train-time image is stored during train-time and prior to said coarse defect detection operation, and wherein said run-time image is masked according to said don'"'"'t care mask to select information required in generating said run-time intensity histogram and said run-time spatial histogram.

9. The method of claim 6 wherein a selected offset is computed for each of said plurality of concentric rings, and said selected offset is added to a starting point of a first sector of each of said plurality of concentric rings to increase angular resolution of said estimated angular displacement determined with respect to said run-time image.

10. The method of claim 6 wherein said step of computing said geometric partition model involves computing a first model and a second model, said second model including at least some overlapping portions with said first model, and said steps of generating a train-time spatial histogram, generating a run-time spatial histogram, and determining an estimated angular displacement of said run-time image relative to said train-time image are performed respectively with respect to each of said first model and said second model and respective information relating to said first model is averaged with respective information relating to said second model prior to comparing said train-time spatial histogram and said aligned run-time spatial histogram to determine a difference therebetween.

11. The method of claim 6 wherein said plurality of sectors comprises 256 sectors, and said train-time spatial histogram and said run-time spatial histogram are each generated in a respective array having a plurality of array elements including a respective array element to store information for each of said plurality of sectors.