Method and apparatus for detecting edges of an object

• US 20060245653A1
• Filed: 03/14/2006
• Published: 11/02/2006
• Est. Priority Date: 03/14/2005
• Status: Active Grant

• Current Assignee: SRI International, Inc.
• Original Assignee: Sarnoff Corporation (SRI International, Inc.)
• Inventors: Hirvonen, David, Camus, Theodore
• Granted Patent: US 7,697,786 B2
• Time in Patent Office: Days
• US Class Current: 382/199
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
• CPC Class Codes: G06V 20/58 Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
• CPC Class Codes: G06V 2201/08 Detecting or categorising vehicles

Abstract

The present invention provides a system and method for detecting edges of an object in a pre-crash sensing environment. The method includes extracting vertical edges from peaks in an array of column sums of absolute values of normalized horizontal derivates computed in the intensity image. The pixels with corresponding range image depths proximate the estimated object'"'"'s closest surface contribute the column sums. The search is conducted for all likely combinations of left/right edge pairs preferably within a pre-determined horizontal range about the object'"'"'s detected center. The left/right edge pairs that delimit a width within the pre-determined range of possible widths for the object are selected. Alternatively, an additional search is performed on each pair of edges to support any angular or inward slant on the edges of the object.

Claims

1. A method for detecting edges of an object, the method comprising:

producing from imagery an image of a scene including said object proximate a platform;

computing horizontal image derivates of the image, wherein the horizontal image derivates are a rate of change in a gradient of the image;

integrating column sums of the horizontal image derivatives to create an edge strength plot of the image;

detecting peaks in the plot corresponding to dominant edges, wherein the dominant edges are potential pairs of left and right edges of the image;

evaluating all the potential pairs of left and right edges of the image; and

selecting the potential pair of left and right edges with maximum score, wherein the maximum score is a score with maximum integrated column sum value.

2. The method of claim 1 further comprising producing from said imagery a depth image of said scene, wherein said depth image determines the horizontal image derivatives to be included in the column sum to be integrated.

3. The method of claim 1 further comprising calculating image coordinates of the selected left and right edges.

4. The method of claim 3 further comprising subtracting the calculated image coordinates to obtain a width of the object.

5. The method of claim 4 further comprising determining top and bottom edge estimates of the object.

6. The method of claim 5 further comprising searching for a pair of left and right edges having image coordinates to be twice the width of the object and bounded by said top and bottom edge estimates.

7. The method of claim 1 further comprising computing and normalizing absolute values of the horizontal derivatives to provide edges with a high gradient.

8. The method of claim 1 wherein the evaluating step includes a pairing algorithm for pairing the left and right edges of the object by combining their individual integrated column sums to provide a paired edge score.

9. The method of claim 8 wherein the potential paired left and right edges include said combinations that fall within a constant depth difference threshold.

10. The method of claim 8 wherein the evaluating step further includes multiplying the edge score by a depth ratio of the potential left and right edges wherein the depth ratio is less than or equal to one.

11. The method of claim 1 wherein the peaks are detected using a mean-shift algorithm.

12. The method of claim 1 wherein the integrating step includes calculating vertical column sums of the horizontal image derivatives.

13. The method of claim 1 wherein the integrating step includes calculating angular column sums of the horizontal image derivates.

14. The method of claim 1 wherein the integrating step includes calculating vertical column sums and angular column sums of the horizontal derivatives.

15. A method for determining a width of the object, the method comprising:

producing from imagery an image of a scene including said object proximate a platform;

computing horizontal image derivates of the image, wherein the horizontal image derivates are a rate of change in a gradient of the image;

integrating column sums of the horizontal image derivatives to create an edge strength plot of the image;

detecting peaks in the plot corresponding to dominant edges, wherein the dominant edges are potential pairs of left and right edges of the image;

evaluating all the potential pairs of left and right edges of the object and selecting the pair of left and right edge with maximum score, wherein the maximum score is a score with maximum integrated column sum value;

calculating image coordinates of the selected pair of the left and right edges; and

subtracting the calculated image coordinates yielding the width of the object.

16. The method of claim 15 further comprising producing from said imagery a depth image of said scene, wherein said depth image determines the horizontal image derivatives to be included in the column sum to be integrated.

17. The method of claim 15 further comprising determining top and bottom edge estimates of the object.

18. The method of claim 15 further comprising searching for a pair of left and right edges having image coordinates to be twice the width of the object and bounded by said top and bottom edge estimates.

19. The method of claim 15 further comprising computing and normalizing absolute values of the horizontal derivatives to provide edges with a high gradient.

20. The method of claim 15 wherein the evaluating step includes a pairing algorithm for pairing the left and right edges of the object by combining their individual integrated column sums to provide a paired edge score.

21. The method of claim 20 wherein the potential paired left and right edges include said combinations that fall within a constant depth difference threshold.

22. The method of claim 20 wherein the evaluating step further includes multiplying the edge score by a depth ratio of the potential left and right edges wherein the depth ratio is less than or equal to one.

23. The method of claim 15 wherein the peaks are detected using a mean-shift algorithm.

24. The method of claim 15 wherein the integrating step includes calculating vertical column sums of the horizontal image derivatives.

25. The method of claim 15 wherein the integrating step includes calculating angular column sums of the horizontal image derivates.

26. The method of claim 15 wherein the integrating step includes calculating vertical column sums and angular column sums of the horizontal derivatives.

27. A pre-crash sensing system, comprising:

a stereo imaging device for producing an image of a scene including an object proximate a platform; and

a dimensions processor for determining edges of an object from said image;

wherein said dimensions processor computes horizontal image derivates of the depth image, the horizontal image derivates are a rate of change in a gradient of the object;

wherein said dimensions processor integrates column sums of the horizontal image derivatives to create an edge strength plot of the object;

wherein said dimensions processor detects peaks in the plot corresponding to dominant edges, the dominant edges are potential pairs of left and right edges of the object; and

wherein said dimensions processor evaluates all the potential pairs of left and right edges of the object and selects the pair of left and right edge with maximum score, wherein the maximum score is a score with maximum integrated column sum value.

28. The system of claim 27 further comprises a depth map generator for producing a depth stereo of the scene, wherein said depth image determines the horizontal image derivates to be included in the column sum to be integrated.

29. The system of claim 27 further comprises a stereo image preprocessor for preprocessing said imagery.

30. The system of claim 27 wherein the dimensions processor calculates image coordinates of the selected pair of left and right edges and subtracts said image coordinates to yield a width of the object.

31. The system of claim 27 wherein said dimensions processor integrates said column sums by calculating vertical column sums of the horizontal image derivates.

32. The system of claim 27 wherein said dimensions processor integrates said column sums by calculating angular column sums of the horizontal image derivates.

33. The system of claim 27 wherein said dimensions processor integrates said column sums by calculating vertical columns sums and angular column sums of the horizontal image derivates.

34. A computer readable medium having stored thereon a plurality of instructions, the plurality of instructions, the plurality of instructions including instructions which, when executed by a dimensions processor, cause the processor to perform the steps of a method for determining edges of an object in an image of a scene including the object, comprising:

computing horizontal image derivates of the image of an object, wherein the horizontal image derivates are a rate of change in a gradient of the image;

integrating column sums of the horizontal image derivatives to create an edge strength plot of the image;

detecting peaks in the plot corresponding to dominant edges, wherein the dominant edges are potential pairs of left and right edges of the image; and

evaluating all the potential pairs of left and right edges of the image and selecting the pair of left and right edge with maximum score, wherein the maximum score is a score with maximum integrated column sum value.