Method for front matching stereo vision

• US 20100007720A1
• Filed: 06/26/2009
• Published: 01/14/2010
• Est. Priority Date: 06/27/2008
• Status: Active Grant

• Current Assignee: University of Southern Mississippi
• Original Assignee: University of Southern Mississippi
• Inventors: Murali, Beddhu
• Granted Patent: US 8,264,526 B2
• Time in Patent Office: Days
• US Class Current: 348/47
• CPC Class Codes: G06T 7/97 Determining parameters from multiple pictures depth or shape recovery from multiple images G06T7/55; stereo camera calibration G06T7/85

Abstract

The present invention provides a method, system, and apparatus for solving the stereo vision matching problem. The invention implements a method for solving image correspondence problems so that each front region between consecutive extremal pixels in one image is matched to a corresponding front region between consecutive extremal pixels in another image.

Claims

1. A method for feature-front based stereo matching image processing utilizing a computer-based device, the method comprising:

providing at least one left image having at least one scanline and at least one right image having at least one scanline of an object or scene, wherein each scanline has at least one extremal pixel;

optionally converting the left image and right image to grey scale images;

optionally smoothing the left and right image;

finding the extremal pixels along each scanline in each image by calculating at each pixel the product of the along-the-scanline backward and forward differences of the image intensity field,

whereby a pixel is an extremal pixel when the product is less than zero or when the product is zero and either the backward or forward difference is non-zero and

wherein two consecutive image intensity extrema are connected by a straight line to form an ascending or descending feature front, for the left image and for the right image, having a strength equal to the absolute difference between the feature front end intensity values,

utilizing the scanlines from the left image and right image simultaneously and sorting the feature fronts in decreasing order of strengths greater than a value of zero into a list; and

calculating for each front in the sorted list, in order of strongest to weakest feature front regardless of scanline, the matching front on the other scanline within a search window satisfying the match criterion, wherein the feature front for which a match is sought is the current feature front.

2. The method of claim 1, wherein the at least one left image and right image are from two cameras having parallel optical axes.

3. The method of claim 1, wherein the at least one left image and right image are from one camera moving parallel to the image plane along the scanline.

4. The method of claim 1, wherein the images are color images.

5. The method of claim 1, wherein the images are raw images.

6. The method of claim 1, wherein the images are smoothed images.

7. The method of claim 1, further comprising finding an initial match on the right image scanline, for the current feature front on the left image scanline, comprising the steps of:

finding the right-limit of the search window as the second nearest front to the right of the current feature front on the left image scanline for which an initial right match was found;

finding the left-limit of the search window as the right end of the extremal interval on the right image scanline in which the current feature front on the left image scanline falls, wherein an extremal interval comprises of all the pixels between consecutive extrema;

computing the sum of the absolute differences between;

the end intensity values of the front on the right image scanline within the search window, of the same type as the current feature front, wherein the type is either ascending or descending, and the current feature front on the left scanline, wherein the tentative right match is the right front minimizing the sum;

similarly searching the left image scanline for a tentative left match of the tentative right match;

accepting the tentative right match as the initial right match for the current feature front when the tentative left match is the same as the current feature front; and

rejecting the tentative right match if the tentative left match is not the same as the current feature front and choosing the next available unmatched and strongest front as the current feature front.

8. The method of claim 1, further comprising finding an initial match on the left image scanline, for the current feature front on the right image scanline, comprising the steps of:

finding the left-limit of the search window as the second nearest front to the left of the current feature front on the right image scanline for which an initial left match was found;

finding the right-limit of the search window as the left end of the extremal interval on the left image scanline in which the current feature front on the right image scanline falls;

computing the sum of the absolute differences between;

the end intensity values of the front on the left image scanline within the search window, of the same type as the current feature front, wherein the type is either ascending or descending, and the current feature front on the right scanline, wherein the tentative left match is the left front minimizing the sum;

similarly searching the right image scanline for a tentative right match of the tentative left match;

accepting the tentative left match as the initial left match for the current feature front when the tentative right match is the same as the current feature front; and

rejecting the tentative left match if the tentative right match is not the same as the current feature front and choosing the next available unmatched and strongest front as the current feature front.

9. The method of claim 7, wherein finding an initial match on the right image scanline continues until the value of the strength of the current feature front is equal to or less than one.

10. The method of claim 8, wherein finding an initial match on the left image scanline continues until the value of the strength of the current feature front is equal to or less than one.

11. The method of claim 1, further comprising finding final matches on the right image scanline, for the feature fronts on the left image scanline, comprising the steps of:

considering each feature front on the left scanline with a valid initial match in order from left to right, wherein the currently considered feature front is the current front;

finding the right-limit on the left scanline as the right end of the extremal interval on the left scanline where the right end of the initial matching front of the current front on the right scanline falls;

searching between the right end of the current front and the right limit for a front with a valid match such that the right end of its matched front is located to the left of the right end of the initial matched front of the current front;

recalculating the match for the current front according to the method of claim 7, if the search succeeds; and

accepting the initial match as the final match if the search fails.

12. The method of claim 1, further comprising finding final matches on the left image scanline, for the feature fronts on the right image scanline, comprising the steps of:

considering each feature front on the right scanline with a valid initial match in order from left to right, wherein the currently considered feature front is the current front;

finding the left-limit on the right scanline as the left end of the extremal interval on the right scanline where the left end of the initial matching front of the current front on the left scanline falls;

searching between the left end of the current front and the left limit for a front with a valid match such that the left end of its matched front is located to the right of the left end of the initial matched front of the current front;

recalculating the match for the current front according to the method of claim 8, if the search succeeds; and

accepting the initial match as the final match if the search fails.

13. The method of claims 11 and 12 wherein the fronts are considered based on their combined strength sorted order.

14. The method of claim 11, further comprising finding two consecutive extremal pixels having valid final matches on the left scanline and interpolating to fill in matches at each in-between pixel.

15. The method of claim 12, further comprising finding two consecutive extremal pixels having valid final matches on the right scanline and interpolating to fill in matches at each in-between pixel.

16. The method of claim 1, further comprising analyzing each scanline of the left image and of the right image for finding valid correspondences for extremal pixels and computing a dense disparity field by interpolating at in-between pixels for computer-based display, storage, or further processing.

17. An apparatus for feature-front based stereo matching image processing utilizing a computer-based device, comprising:

means for providing at least one left image having at least one scanline and at least one right image having at least one scanline of an object or scene, wherein each scanline has at least one extremal pixel;

means for optionally converting the left image and right image to grey scale images;

means for optionally smoothing the left and right image;

means for finding the extremal pixels along each scanline in each image by calculating at each pixel the product of the along-the-scanline backward and forward differences of the image intensity field,

whereby a pixel is an extremal pixel when the product is less than zero or when the product is zero and either the backward or forward difference is non-zero and

wherein two consecutive image intensity extrema are connected by a straight line to form an ascending or descending feature front, for the left image and for the right image, having a strength equal to the absolute difference between the feature front end intensity values;

means for utilizing the scanlines from the left image and right image simultaneously and sorting the feature fronts in decreasing order of strengths greater than a value of zero into a list; and

means for calculating for each front in the sorted list, in order of strongest to weakest feature front regardless of scanline, the matching front on the other scanline within a search window satisfying the match criterion, wherein the feature front for which a match is sought is the current feature front.

18. An apparatus for feature-front based stereo matching image processing as defined in claim 17, wherein the at least one left image and right image are from two cameras having parallel optical axes or from one camera moving parallel to the image plane along the scanline.

19. An apparatus as defined in claim 17, further comprising a means for finding an initial match on the right image scanline, for the current feature front on the left image scanline.

20. An apparatus as defined in claim 17, further comprising a means for finding an initial match on the left image scanline, for the current feature front on the right image scanline.

21. An apparatus as defined in claim 19, wherein the means for finding an initial match on the right image scanline continues until the value of the strength of the current feature front is equal to or less than one.

22. An apparatus as defined in claim 20, wherein the means for finding an initial match on the left image scanline continues until the value of the strength of the current feature front is equal to or less than one.

23. An apparatus as defined in claim 17, further comprising a means for finding final matches on the right image scanline, for the feature fronts on the left image scanline.

24. An apparatus as defined in claim 17, further comprising a means for finding final matches on the left image scanline, for the feature fronts on the right image scanline.

25. An apparatus as defined in claims 23 and 24, wherein the fronts are considered based on their combined strength sorted order.

26. An apparatus as defined in claim 23, further comprising a means for finding two consecutive extremal pixels having valid final matches on the left scanline and interpolating to fill in matches at each in-between pixel.

27. An apparatus as defined in claim 24, further comprising a means for finding two consecutive extremal pixels having valid final matches on the right scanline and interpolating to fill in matches at each in-between pixel.

28. An apparatus as defined in claim 17, further comprising a means for analyzing each scanline of the left image and of the right image for finding valid correspondences for extremal pixels and a means for computing a dense disparity field by interpolating at in-between pixels for computer-based display, storage, or further processing.

29. An apparatus as defined in claim 17, wherein the images are color images.

30. An apparatus as defined in claim 17, wherein the images are raw images.

31. An apparatus as defined in claim 17, wherein the images are smoothed images.