Method and System for Dynamic Feature Detection

• US 20090257621A1
• Filed: 04/09/2008
• Published: 10/15/2009
• Est. Priority Date: 04/09/2008
• Status: Active Grant

• Current Assignee: Cognex Corporation
• Original Assignee: Cognex Corporation
• Inventors: Silver, William M.
• Granted Patent: US 8,238,639 B2
• Time in Patent Office: Days
• US Class Current: 382/103
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 7/246 using feature-based methods, e.g. the tracking of corners or segments
• CPC Class Codes: G06V 10/62 relating to a temporal dimension, e.g. time-based feature extraction; Pattern tracking
• CPC Class Codes: G06V 10/751 Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching

Abstract

Disclosed are methods and systems for dynamic feature detection of physical features of objects in the field of view of a sensor. Dynamic feature detection substantially reduces the effects of accidental alignment of physical features with the pixel grid of a digital image by using the relative motion of objects or material in and/or through the field of view to capture and process a plurality of images that correspond to a plurality of alignments. Estimates of the position, weight, and other attributes of a feature are based on an analysis of the appearance of the feature as it moves in the field of view and appears at a plurality of pixel grid alignments. The resulting reliability and accuracy is superior to prior art static feature detection systems and methods.

Claims

1. A method for measuring a plurality of object features on an object, comprising:

providing relative motion between the object and a field of view of a sensor;

capturing a sequence of images of the field of view;

detecting, for each image of the sequence of images, a set of static features;

storing in a memory a plurality of dynamic features corresponding to the plurality of object features, each dynamic feature comprising at least one measurement of a physical property of the corresponding object feature;

analyzing, for each image of the sequence of images, the set of static features and the stored plurality of dynamic features so as to determine a map, the map comprising

a pose that maps between points on the object and points in the image, the pose comprising at least one degree of freedom; and

an association that maps dynamic features to corresponding static features; and

updating at least one dynamic feature, responsive to the map, so as to improve at least one measurement of a physical property of the corresponding object feature.

2. The method of claim 1, wherein the object is of substantially unknown appearance, and further comprising creating and storing in the memory new members of the plurality of dynamic features in response to the relative motion between the object and the field of view.

3. The method of claim 2, further comprising removing dynamic features from the memory in response to the relative motion between the object and the field of view.

4. The method of claim 2, wherein

each dynamic feature further comprises an experience factor;

the analyzing is responsive to the experience factors of the dynamic features, such that at least one dynamic feature influences the determined map to an extent that increases as its experience factor increases; and

the updating is further responsive to the experience factors of the dynamic features, such that at least one dynamic feature is influenced by the determined map to an extent that decreases as its experience factor increases.

5. The method of claim 4, wherein the experience factor is responsive to a count of the images that have been captured since the dynamic feature was created.

6. The method of claim 4, wherein

each static feature comprises a weight indicating relative confidence that the static feature is responsive to an object feature;

and

the experience factor is responsive to the weights of the static features to which the dynamic feature corresponds.

7. The method of claim 1, wherein the object is of substantially known appearance and the stored plurality of dynamic features are predetermined responsive to the known appearance.

8. The method of claim 7, further comprising creating the predetermined plurality of dynamic features responsive to a training object placed in the field of view.

9. The method of claim 8, wherein creating the predetermined plurality of dynamic features comprises

providing relative motion between the training object and the field of view;

capturing a second sequence of images of the field of view;

detecting, for each image of the second sequence of images, a set of training static features;

storing in a memory a plurality of training dynamic features, each training dynamic feature comprising at least one measurement of a physical property of a corresponding training object feature, and wherein the plurality of training dynamic features are not predetermined;

analyzing, for each image of the second sequence of images, the corresponding set of training static features and the stored plurality of training dynamic features so as to determine a training map, the training map comprising

a training pose that maps between points on the training object and points in the image; and

a training association that maps training dynamic features to corresponding training static features; and

updating at least one training dynamic feature, responsive to the training map, so as to improve at least one measurement of the physical property of the corresponding training object feature; and

creating the predetermined plurality of dynamic features from the plurality of training dynamic features.

10. The method of claim 7, further comprising

examining the stored plurality of dynamic features so as to determine at least one score indicating relative confidence that an object substantially similar in appearance to the known appearance is in the field of view;

using the at least one score to make a decision on whether an object substantially similar in appearance to the known appearance has been detected; and

outputting a detection signal responsive to the decision.

11. The method of any of claims 1, 2, 4, and 7, further comprising outputting a motion signal comprising information about the motion of the object, the motion signal responsive to at least one degree of freedom of the pose.

12. A system for measuring a plurality of object features on an object, comprising:

a sensor having a field of view;

a motion process that provides relative motion between the object and the field of view of the sensor;

a capture process that captures a sequence of images of the field of view;

a static feature detection process that detects, for each image of the sequence of images, a set of static features;

a memory that stores a plurality of dynamic features corresponding to the plurality of object features, each dynamic feature comprising at least one measurement of a physical property of the corresponding object feature;

an analysis process that analyzes, for each image of the sequence of images, the set of static features and the stored plurality of dynamic features so as to determine a map, the map comprising

a pose that maps between points on the object and points in the image, the pose comprising at least one degree of freedom; and

an association that maps dynamic features to corresponding static features; and

a learning process that updates at least one dynamic feature, responsive to the map, so as to improve at least one measurement of a physical property of the corresponding object feature.

13. The system of claim 12, wherein the object is of substantially unknown appearance, and further comprising a birth process that creates and stores in the memory new members of the plurality of dynamic features in response to the relative motion between the object and the field of view.

14. The system of claim 13, further comprising a death process that removes dynamic features from the memory in response to the relative motion between the object and the field of view.

15. The system of claim 13, wherein

each dynamic feature further comprises an experience factor;

the analysis process is responsive to the experience factors of the dynamic features, such that at least one dynamic feature influences the determined map to an extent that increases as its experience factor increases; and

the learning process is further responsive to the experience factors of the dynamic features, such that at least one dynamic feature is influenced by the determined map to an extent that decreases as its experience factor increases.

16. The system of claim 15, wherein the experience factor is responsive to a count of the images that have been captured since the dynamic feature was created.

17. The system of claim 15, wherein

each static feature comprises a weight indicating relative confidence that the static feature is responsive to an object feature;

and

the experience factor is responsive to the weights of the static features to which the dynamic feature corresponds.

18. The system of claim 12, wherein the object is of substantially known appearance and the stored plurality of dynamic features are predetermined responsive to the known appearance.

19. The system of claim 18, further comprising a training process that creates the predetermined plurality of dynamic features responsive to a training object placed in the field of view.

20. The system of claim 19, wherein the training process comprises

a training motion process that provides relative motion between the training object and the field of view;

a training capture process that captures a second sequence of images of the field of view;

a training static feature detection process that detects, for each image of the second sequence of images, a set of training static features;

a training memory that stores a plurality of training dynamic features, each training dynamic feature comprising at least one measurement of a physical property of a corresponding training object feature, and wherein the plurality of training dynamic features are not predetermined;

a training analysis process that analyzes, for each image of the second sequence of images, the corresponding set of training static features and the stored plurality of training dynamic features so as to determine a training map, the training map comprising

a training pose that maps between points on the training object and points in the image; and

a training association that maps training dynamic features to corresponding training static features; and

a training learning process that updates at least one training dynamic feature, responsive to the training map, so as to improve at least one measurement of the physical property of the corresponding training object feature; and

a selection process that creates the predetermined plurality of dynamic features from the plurality of training dynamic features.

21. The system of claim 18, further comprising

a detection process that examines the stored plurality of dynamic features so as to determine at least one score indicating relative confidence that an object substantially similar in appearance to the known appearance is in the field of view;

a decision process that uses the at least one score to make a decision on whether an object substantially similar in appearance to the known appearance has been detected; and

a signaling process that outputs a detection signal responsive to the decision.

22. The system of any of claims 12, 13, 15, and 18, further comprising a signaling process that outputs a motion signal comprising information about the motion of the object, the motion signal responsive to at least one degree of freedom of the pose.