System and methods for automatic parameter determination in machine vision

• US 7,751,625 B2
• Filed: 04/21/2006
• Issued: 07/06/2010
• Est. Priority Date: 03/24/2006
• Status: Active Grant

• Current Assignee: MvTec Software GmbH (MvTec LLC)
• Original Assignee: MvTec Software GmbH (MvTec LLC)
• Inventors: Ulrich, Markus, Steger, Carsten
• Primary Examiner(s): COUSO, YON JUNG
• Application Number:
  US11/409,531
• Publication Number:
  US 20070223817A1
• Time in Patent Office: 1,537 Days
• Field of Search: 382/199, 382/141, 382/144, 382/148, 382/151, 382/152, 382/161, 382/272, 382/286
• US Class Current: 382/199
• CPC Class Codes: G06F 18/28 Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
• CPC Class Codes: G06V 10/7515 Shifting the patterns to accommodate for positional errors
• CPC Class Codes: G06V 10/772 Determining representative reference patterns, e.g. averaging or distorting patterns; Generating dictionaries

Abstract

The present invention provides a system and methods for automatic parameter determination in machine vision in general, and in object recognition in particular. Many machine vision systems use algorithms that demand the user to specify one or more parameters in order to adapt the behavior of the algorithm in dependence of the current application. This is not desirable because the complexity of the algorithm should be hidden from the user and a manual parameter determination is contrary to a desirable high degree of automation. The present invention provides a method to automatically determine the most frequently used parameters in machine vision solely based on the input image itself. The method is explained in detail using an object recognition system as an example. In particular, the model generation process based on a model image of the object is explained. However, also other systems that use edge extraction algorithms, for example, can benefit from the present invention.

Claims

1. A method for automatically determining a discretization step length of an object model transformation comprising a set of model points, the method comprising the steps of:

(a) providing a digital electronic image of the object with a camera;

(b) providing an initial value for the discretization step length based on the size of the object;

(c) identifying a minimum similarity measure that is associated with the current discretization step;

(d) halving the discretization step length until the associated minimum similarity measure exceeds a certain threshold to determine a resultant discretization step; and

(e) using the resultant discretization step length to generate model transformations for object recognition.

2. The method of claim 1, wherein step (b) comprises the steps of

(b1) computing the maximum distance of the model points from the fixed point of the transformation;

(b2) assuming a maximum allowed pixel displacement;

(b3) computing the initial value for the discretization step length based on the maximum distance and the allowed pixel displacement.

3. The method of claim 1, wherein the minimum similarity measure in step (c) corresponds to the similarity measure between a non-transformed object model and the model image that has been transformed by half the discretization step length.

4. A method for automatically determining a discretization step length of an object model comprising the steps of:

(a) providing an electronic image of the object;

(b) initializing a current step length with 0;

(c) incrementing the current step length by a value that corresponds to the accuracy with which the discretization step length is to be determined;

(d) identifying a similarity measure between a non-transformed model and a model image transformed in accordance with the current step length;

(e) repeating the steps (c) and (d) until the similarity measure is below a certain threshold; and

(f) providing an optimum step length as twice the last step length for which the similarity measure is above the threshold.

5. A data processing system for automatically determining a discretization step length of an object model, the system comprising:

(a) a means for providing an electronic image of the object; and

(b) a means for providing an initial value for the discretization step length based on the size of the object;

identifying a minimum similarity measure that is associated with the current discretization step;

halving the discretization step length until the associated minimum similarity measure exceeds a certain threshold to determine a resultant discretization step; and

using the resultant discretization step length to determine model transformations for object recognition.

6. A computer-readable medium including program instructions for automatically determining a discretization step length of an object model, wherein the program steps comprise:

(a) providing an electronic image of the object;

(b) providing an initial value for the discretization step length based on the size of the object;

(c) identifying a minimum similarity measure that is associated with the current discretization step;

(d) halving the discretization step length until the associated minimum similarity measure exceeds a certain threshold to determine a resultant discretization step; and

(e) using the resultant discretization step length to determine model transformations for object recognition.