System and methods for rogue can detection

• US 20060204080A1
• Filed: 03/10/2005
• Published: 09/14/2006
• Est. Priority Date: 03/10/2005
• Status: Active Grant

• Current Assignee: Applied Vision Company, LLC
• Original Assignee: Applied Vision Company, LLC
• Inventors: Baird, Brain M., Sones, Richard A.
• Granted Patent: US 7,308,142 B2
• Time in Patent Office: Days
• US Class Current: 382/155
• CPC Class Codes: G06T 7/0004 Industrial image inspection

Abstract

A vision system and methods to detect rogue objects on a product processing line are disclosed. The vision system includes a camera, a source of illumination, a user interface, and a computer-based platform including a memory. The vision system captures images from objects on the product processing line as the objects pass by the camera which is mounted on the product processing line. Image pixel data is converted to vectors of mean pixel values for each captured image from each object, using the vision system. The vectors are compared to a vector training set stored within the memory of the vision system to determine if the objects correspond to “rogue” objects or not, using the vision system. The vector training set is representative of the current product (i.e., objects) being processed on the product processing line.

Claims

1. A method of training a vision system on a process monitoring line, said method comprising:

acquiring one image from each object of a predetermined number of objects passing by a vision system on a process monitoring line using said vision system, wherein each said image corresponds to a random spatial orientation of each said corresponding object with respect to said vision system, and wherein each said image comprises at least one array of pixel values stored within said vision system;

generating a plurality of mean pixel values, along one dimension of said at least one array of pixel values, for each said acquired image to form a vector of mean pixel values for each said acquired image, using said vision system; and

storing each said vector of mean pixel values for each said acquired image within a memory of said vision system to form a vector training set of said vision system.

2. The method of claim 1 wherein said one dimension of said at least one array of pixel values comprises a horizontal row dimension.

3. The method of claim 1 wherein said one dimension of said at least one array of pixel values comprises a vertical column dimension.

4. The method of claim 1 wherein said at least one array of pixel values comprises an array of red (R) color pixel values, an array of green (G) color pixel values, and an array of blue (B) color pixel values for each said acquired image.

5. The method of claim 4 wherein each said vector of mean pixel values comprises a sub-vector of mean red pixel values, a sub-vector of mean green pixel values, and a sub-vector of mean blue pixel values.

6. The method of claim 1 further comprising:

acquiring a first next image from a first next object passing by said vision system on said process monitoring line immediately after said predetermined number of objects, using said vision system, wherein said first next image corresponds to a random spatial orientation of said first next object with respect to said vision system, and wherein said first next image comprises at least one array of pixel values stored within said vision system;

generating a first next plurality of mean pixel values, along said one dimension of said at least one array of pixel values of said first next image, to form a first next vector of mean pixel values using said vision system; and

replacing a first stored vector of mean pixel values of said vector training set within said memory of said vision system with said first next vector of mean pixel values.

7. The method of claim 6 further comprising:

acquiring a second next image from a second next object passing by said vision system on said process monitoring line immediately after said first next object, using said vision system, wherein said second next image corresponds to a random spatial orientation of said second next object with respect to said vision system, and wherein said second next image comprises at least one array of pixel values stored within said vision system;

generating a second next plurality of mean pixel values, along said one dimension of said at least one array of pixel values of said second next image, to form a second next vector of mean pixel values, using said vision system; and

replacing a second stored vector of mean pixel values of said vector training set within said memory of said vision system with said second next vector of mean pixel values.

8. The method of claim 7 wherein said one dimension comprises a horizontal row dimension.

9. The method of claim 7 wherein said one dimension comprises a vertical column dimension.

10. The method of claim 7 wherein said at least one array of pixel values for each said acquired image comprises at least one array of red, green, and blue (RGB) color pixel values.

11. The method of claim 10 wherein each said vector of mean pixel values comprises a sub-vector of mean red pixel values, a sub-vector of mean green pixel values, and a sub-vector of mean blue pixel values.

12. A method of monitoring objects on a process monitoring line, said method comprising:

acquiring an image from a next object passing by a vision system on a process monitoring line, using said vision system, wherein said image corresponds to a random spatial orientation of said next object with respect to said vision system, and wherein said image comprises at least one array of pixel values stored within said vision system;

generating a plurality of mean pixel values, along one dimension of said at least one array of pixel values for said acquired image, to form a next vector of mean pixel values for said acquired image, using said vision system;

comparing said next vector of mean pixel values to each training vector of mean pixel values in a stored vector training set in said vision system to generate one discrepancy value for each training vector in said vector training set, forming a set of discrepancy values within said vision system; and

selecting one discrepancy value from said set of discrepancy values and comparing said selected discrepancy value to a predetermined threshold value to determine if said next object is a rogue object, using said vision system.

13. The method of claim 12 further comprising declaring said next object to be a rogue object if said selected discrepancy value is greater than said predetermined threshold value, using said vision system.

14. The method of claim 12 wherein said selected discrepancy value corresponds to a smallest discrepancy value of said set of discrepancy values.

15. The method of claim 12 further comprising replacing a next vector of said vector training set with said next vector of mean pixel values for said acquired image if said next object is not a rogue object, using said vision system.

16. The method of claim 12 wherein said one dimension of said at least one array of pixel values comprises a horizontal row dimension.

17. The method of claim 12 wherein said one dimension of said at least one array of pixel values comprises a vertical column dimension.

18. The method of claim 12 wherein said at least one array of pixel values comprises at least one array of red, green, and blue (RGB) color pixel values for said acquired image.

19. The method of claim 18 wherein said next vector of mean pixel values comprises a sub-vector of mean red pixel values, a sub-vector of mean green pixel values, and a sub-vector of mean blue pixel values.

20. A vision system for training and monitoring a process, said vision system comprising:

a source of illumination positioned to illuminate objects as said objects move along a process monitoring line in spatially random orientations;

a camera positioned on said process monitoring line to capture at least one image from each of said illuminated objects, forming a plurality of images, as each object passes through a field-of-view of said camera; and

a computer-based platform being connected to said camera to generate a vector of mean pixel values from each of said plurality of images, forming a plurality of vectors of mean pixel values, and to store at least some of said vectors of mean pixel values as a vector training set.

21. The vision system of claim 20 wherein said computer-based platform includes a frame grabber to convert analog signals, output from said camera, to digital signals representing single frames of digital pixel values.

22. The vision system of claim 20 wherein said plurality of images correspond to random rotational positions of said plurality of objects with respect to said camera as said objects move along said process monitoring line past said camera.

23. The vision system of claim 20 wherein said source of illumination comprises an array of light emitting diodes which emits a spectrum of white light.

24. The vision system of claim 20 wherein said camera outputs analog imaging signals.

25. The vision system of claim 20 wherein said camera outputs digital imaging signals.

26. The vision system of claim 20 further comprising a user interface for at least controlling said vision system.

27. A method for training and monitoring a process, said method comprising:

generating a vector training set, comprising training vectors of mean pixel values, from images acquired from a predetermined number of objects passing by a vision system on a process monitoring line, using said vision system;

generating a first next vector of mean pixel values from a first next image acquired from a first next object passing by said vision system on said process monitoring line immediately after said predetermined number of objects, using said vision system;

comparing said first next vector to each training vector of said vector training set to generate one discrepancy value for each training vector, forming a set of discrepancy values, using said vision system; and

selecting one discrepancy value from said set of discrepancy values and comparing said selected discrepancy value to a predetermined threshold value to determine if said next object is a rogue object, using said vision system.

28. The method of claim 27 further comprising replacing a first training vector of said vector training set with said first next vector if said first next object is not a rogue object, using said vision system.

29. The method of claim 28 further comprising:

generating a second next vector of mean pixel values from a second next image acquired from a second next object passing by said vision system on said process monitoring line immediately after said first next object, using said vision system;

comparing said second next vector to each training vector of said vector training set to generate one discrepancy value for each training vector, forming a next set of discrepancy values, using said vision system; and

selecting one next discrepancy value from said next set of discrepancy values and comparing said selected next discrepancy value to said predetermined threshold value, using said vision system.

30. The method of claim 29 further comprising replacing a second training vector of said vector training set with said second next vector if said second next object is not a rogue object, using said vision system.

31. The method of claim 27 further comprising declaring said first next object to be a rogue object if said selected discrepancy value is greater than said predetermined threshold value, using said vision system.

32. The method of claim 29 further comprising declaring said second next object to be a rogue object if said selected next discrepancy value is greater than said predetermined threshold value, using said vision system.