Method and apparatus for training a probe model based machine vision system
First Claim
1. A method for training a pattern recognition algorithm for a machine vision system that uses models of a pattern to be located, the method comprising the steps of:
- training each of a plurality of models using a different training image wherein each of the training images is a version of at least one image of the pattern at a unique coarse image resolution, wherein each model includes a plurality of probes that are each a measure of similarity of at least one of a run-time image feature and a run time image region to at least one of a pattern feature and a pattern region, respectively, at a specific location;
using the models to identify at least one robust image resolution where the robust image resolution is suitable for locating the pattern within an accuracy limit of the actual location of the pattern in the at least one image of the pattern;
storing the at least one robust image resolution for use in subsequent pattern recognition processes during run time image processing;
applying the plurality of probes at a plurality of poses to a run-time image resulting in a plurality of score spaces at each pose where the score spaces include score-space peaks, the score-space peaks indicating likely locations of occurrences of the pattern in the run-time image at each pose;
and comparing the score-space peaks to an accept threshold where the results of the comparison are used to identify poses of instances of the model in the image.
1 Assignment
0 Petitions
Accused Products
Abstract
A method for training a pattern recognition algorithm for a machine vision system that uses models of a pattern to be located, the method comprising the steps of training each of a plurality of models using a different training image wherein each of the training images is a version of a single image of the pattern at a unique coarse image resolution, using the models to identify at least one robust image resolution where the image resolution is suitable for locating the pattern within an accuracy limit of the actual location of the pattern in the image and storing the at least one robust image resolution for use in subsequent pattern recognition processes.
-
Citations
41 Claims
-
1. A method for training a pattern recognition algorithm for a machine vision system that uses models of a pattern to be located, the method comprising the steps of:
-
training each of a plurality of models using a different training image wherein each of the training images is a version of at least one image of the pattern at a unique coarse image resolution, wherein each model includes a plurality of probes that are each a measure of similarity of at least one of a run-time image feature and a run time image region to at least one of a pattern feature and a pattern region, respectively, at a specific location; using the models to identify at least one robust image resolution where the robust image resolution is suitable for locating the pattern within an accuracy limit of the actual location of the pattern in the at least one image of the pattern; storing the at least one robust image resolution for use in subsequent pattern recognition processes during run time image processing; applying the plurality of probes at a plurality of poses to a run-time image resulting in a plurality of score spaces at each pose where the score spaces include score-space peaks, the score-space peaks indicating likely locations of occurrences of the pattern in the run-time image at each pose; and comparing the score-space peaks to an accept threshold where the results of the comparison are used to identify poses of instances of the model in the image. - View Dependent Claims (2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19)
-
-
20. A method for training a pattern recognition algorithm for a machine vision system that uses models of a pattern to be located wherein the pattern includes repeating elements, the method comprising the steps of:
-
training each of a plurality of models using a different training image wherein each of the training images is a version of at least one image of the pattern at a unique coarse image resolution; using the models to identify at least one robust image resolution where the robust image resolution is suitable for locating the pattern within a fraction of one pitch of the repeating elements; storing the at least one robust image resolution for use in subsequent pattern recognition processes during run time image processing; wherein the steps of training, using and storing include, starting with a coarse image having a coarse image resolution where the coarse image resolution is a test image resolution; (a) using the image at the test image resolution to train a model; (b) running the model on the image at the test image resolution to generate a score space and a location error; (c) using the location error to determine if the test image resolution is suitable for locating the pattern within a fraction of one pitch of the repeating elements; (d) where the test image resolution is suitable for locating the pattern within a fraction of one pitch of the repeating elements, using the score space to identify a sparsest robust scan code for the test image resolution where the test image resolution and sparsest robust scan code comprise a test image resolution and scan code combination; (e) storing the test image resolution and scan code combination for use in subsequent pattern recognition processes as a stored image resolution/scan code combination; (f) determining if a less coarse image resolution and scan code combination could be faster than the stored image resolution/scan code combination; (g) where a less coarse image resolution and scan code combination cannot be faster than the stored image resolution/scan code combination, ending the process; and (h) where a less coarse image resolution and scan code combination could be faster than the stored image resolution/scan code combination, repeating steps (a) through (g) using a less coarse test image resolution and only scan codes that could be faster at the less coarse test image resolution than the stored image resolution/scan code combination. - View Dependent Claims (21, 22, 23, 24)
-
-
25. An apparatus for training a pattern recognition algorithm for a machine vision system that uses models of a pattern to be located, the apparatus comprising:
-
a processor programmed to perform the steps of;
training each of a plurality of models using a different training image wherein each of the training images is a version of a single image of the pattern at a unique coarse image resolution, each model including a plurality of probes that are each a measure of similarity of at least one of a run-time image feature and a run time image region to at least one of a pattern feature and a pattern region, respectively, at a specific location;using the models to identify at least one robust image resolution where the image resolution is suitable for locating the pattern within an accuracy limit of the actual location of the pattern in the single image of the pattern; and storing the at least one robust image resolution for use in subsequent pattern recognition processes during run time image processing; applying the plurality of probes at a plurality of poses to a run-time image resulting in a plurality of score spaces at each pose where the score spaces include score-space peaks, the score-space peaks indicating likely locations of occurrences of the pattern in the run-time image at each pose; and comparing the score-space peaks to an accept threshold where the results of the comparison are used to identify poses of instances of the model in the image. - View Dependent Claims (26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39)
-
-
40. The apparatus for training a pattern recognition algorithm for a machine vision system that uses models of a pattern to be located wherein the pattern includes repeating elements, the apparatus comprising:
-
a processor programmed to perform the steps of; training each of a plurality of models using a different training image wherein each of the training images is a version of a single image of the pattern at a unique coarse image resolution; using the models to identify at least one robust image resolution where the image resolution is suitable for locating the pattern within a fraction of one pitch of the repeating elements in the single image of the pattern; storing the at least one robust image resolution for use in subsequent pattern recognition processes during run time image processing; wherein the processor is further programmed to perform the step of identifying a fastest image resolution and scan code combination where the image resolution is suitable for locating the pattern within the accuracy limit of the actual location of the pattern in the single image and the scan code is robust at the image resolution in the image resolution and scan code combination, the processor programmed to perform the step of storing the at least one robust image resolution by storing at least one robust image resolution and scan code combination; and wherein the processor is programmed to perform the steps of training, using and storing by, starting with a coarse image having a coarse image resolution where the coarse image resolution is a test image resolution; (a) using the image at the test image resolution to train a model; (b) running the model on the image at the test image resolution to generate a score space and a location error; (c) using the location error to determine if the test image resolution is suitable for locating the pattern within a fraction of one pitch of the repeating elements; (d) where the test image resolution is suitable for locating the pattern within a fraction of one pitch of the repeating elements, using the score space to identify a sparsest robust scan code for the test image resolution where the test image resolution and sparsest robust scan code comprise a test image resolution and scan code combination; (e) storing the test image resolution and scan code combination for use in subsequent pattern recognition processes as a stored image resolution/scan code combination; (f) determining if a less coarse image resolution and scan code combination could be faster than the stored image resolution/scan code combination; (g) where a less coarse image resolution and scan code combination cannot be faster than the stored image resolution/scan code combination, ending the process; and (h) where a less coarse image resolution and scan code combination could be faster than the stored image resolution/scan code combination, repeating steps (a) through (g) using a less coarse test image resolution and only scan codes that could be faster at the less coarse test image resolution than the stored image resolution/scan code combination. - View Dependent Claims (41)
-
Specification