Method and apparatus for image registration

• US 6,504,957 B2
• Filed: 02/27/2002
• Issued: 01/07/2003
• Est. Priority Date: 07/07/1997
• Status: Expired due to Term

• Current Assignee: General Electric Company
• Original Assignee: General Electric Company
• Inventors: Nguyen, Van-Duc, Nzomigni, Victor, Stewart, Charles Vernon
• Primary Examiner(s): Couso, Jose L.
• Assistant Examiner(s): DESIRE, GREGORY M
• Application Number:
  US10/085,158
• Publication Number:
  US 20020118892A1
• Time in Patent Office: 314 Days
• Field of Search: 382/100, 382/103, 382/106, 382/141, 382/145, 382/149-150, 382/154, 382/209, 382/218, 382/224, 382/270, 382/282, 382/284, 382/291, 382/294, 382/296, 348/36, 348/87, 348/590, 356/237, 356/239, 356/371, 356/394, 345/216, 345/218
• US Class Current: 382/209
• CPC Class Codes: G01N 2021/8887 based on image processing techniques
• CPC Class Codes: G01N 21/9515 Objects of complex shape, e.g. examined with use of a surface follower device measuring contours and curvatures G01B11/24
• CPC Class Codes: G06T 3/073 Transforming surfaces of revolution to planar images, e.g. cylindrical surfaces to planar images
• CPC Class Codes: G06V 10/7515 Shifting the patterns to accommodate for positional errors
• CPC Class Codes: G06V 20/64 Three-dimensional objects
• CPC Class Codes: Y10S 345/953 Geometric processing

Abstract

A machine vision system includes an apparatus for registering an input image of an object, such as an aircraft engine blade, to a reference image comprising ideal specifications for the object in order to detect flaws in the object. The system includes one or more imaging devices for obtaining the input image representing the object. The system further includes a processor for registering the input image to the reference image. The processor includes a patch determining device for identifying low curvature portions of the reference image off-line. A transformation estimator matches the low curvature portions of the reference image to corresponding low curvature portions of the object, and provides a transformation matrix which maps points on the reference image 11 to corresponding points on the input image for precise and fast registration.

Claims

1. A method of comparing a theoretical N dimensional model surface to N dimensional measured data surface points of an object comprising the steps of:

a) acquiring said N dimensional model;

b) analyzing said model to determine a plurality of low curvature surface patches P_i, each P_i having a curvature less than a predetermined curvature threshold;

c) determining a surface normal vector n_i, and center location p_i for each of said low curvature surface patches P_i;

d) projecting a line from p_i to intersect with location q_i on the surface data;

e) determining a pose error, being a sum of distances d(P, P′

) between locations p_i and q_i for a plurality of patches i;

f) minimizing the sum of distances d(P, P′

) to result in the pose error;

g) adjusting the pose of the N dimensional model relative to the N-dimensional data to according to the calculated pose error as a rough pose;

h) calculating a statistical distribution of a plurality of data surface points;

i) determining deviation from the calculated statistical distribution of a plurality of data surface;

j) comparing the each of the plurality of data surface points to the statistical distribution;

k) attributing a weight to each of the plurality of surface data points based upon the comparison to the statistical distribution, being between 1 and 0, with the weight going toward 0 as the distance from the center of the statistical distribution increases.

l) multiplying each range data point by its weighting to result in weighted surface data;

m) repeating steps (b) through (g) above to solve for fine pose using the weighted measured data in place of the original measured data;

n) adjusting the rough pose of the N dimensional model relative to the N-dimensional surface data to according to the calculated pose error from the weighted measured data.

2. A system for adjusting pose registration between a theoretical N dimensional model surface to N dimensional measured data surface represented by the model comprising:

a) a patch determination device for determining a plurality of patches P_i each having less than a predetermined surface curvature;

b) a patch center device functionally coupled to the patch determination device for calculating a center point p_i for each of the plurality of low curvature surface patches;

c) normal vector calculation device functionally coupled to the patch determination device and the patch center device for determining a vector n_i normal to each patch P_i at its center point p_i;

d) a geometric determination device functionally coupled to the patch center, and normal vector calculation devices, and a surface data source, which extends the normal vector n_i until it intersects with surface data at location q_i;

e) a distance minimization device functionally coupled to the patch center, normal vector calculation and geometric calculation devices for determination of rotation intermediates (ω

, t″

);

f) rotation conversion device coupled to the distance minimization device for receiving rotation intermediates (W, t″

) and for converting them into pose error (R, t); and

g) an offset device coupled to the rotation conversion device, the surface data source and the model source, for receiving pose error (R, t) and adjusting the surface data and model pose according to the pose error (R, t).

3. The system for adjusting pose registration of claim 2, wherein the surface data source comprises:

a) a physical object to be matched; and

b) a ranging device which scans the surface of the object to provide surface data.

4. The system for adjusting pose registration of claim 3, further comprising a weighting device functionally coupled to the surface source device, the model source and distance minimization device for determining when a pose error (R, t) has been calculated and for determining a weighting factor which becomes smaller as the distance of the point from the model surface becomes larger, and is a maximum when the surface data point coincides with the model surface, for the plurality of surface points and multiplying each surface point by its weighting factor, then storing the weighted surface points in the memory.