Camera calibration

1. A system for image calibration, comprising:

a processing device; and

modules to be implemented by the processing device, the modules comprising:

a feature module to:

detect features on a set of images;

describe features on the set of images;

determine a match between features of the image sets; and

a homography module to:

determine a shift on the matched features based on camera positions;

determine a first homography between the camera positions and the determined shift;

reconstruct the detected features in three-dimensions; and

determine a second homography based on a re-projection of the three-dimensional features back to a plurality of cameras.

2. The system of claim 1 , further comprising a rectifier module configured to rectify the set of images with an original homography from a manufacture calibration, wherein the rectification is performed before the feature detection of the feature module.

3. The system of claim 1 , wherein the images of an image set comprise images each captured at a different camera comprising at least three cameras.

4. The system of claim 1 , wherein the images of the image set comprise images captured at a single camera disposed at different positions.

5. The system of claim 1 , further comprising an outlier module configured to:

discard outliers based on feature match angle above a predetermined threshold; and

discard outliers based on a feature match size above a predetermined threshold.

6. The system of claim 5 , wherein the outlier module is further configured to discard outliers where a feature match occurs on less than three cameras.

7. The system of claim 5 , wherein the outlier module is further configured to determine a scale of the shift indicating the difference of feature coordinates between image sets divided by a displacement associated with each image captured.

8. The system of claim 1 , wherein the second homography is homography between original feature coordinates from manufacturer calibration and feature coordinates based on the re-projection of the three dimension reconstruction to one or more cameras.

9. The system of claim 8 , wherein the three dimension reconstruction estimates scene depth.

10. The system of claim 1 , wherein the modules are operable ad hoc as images are captured.

11. A method for image calibration, comprising:

detecting, with a processing device, features on a set of images;

describing features on the set of images;

determining a match between features of the image sets;

determining a shift on the matched features based on camera positions associated with the matched features;

determining a first homography between the camera positions and the determined shift; reconstructing the detected features in three-dimensions; and

determining a second homography based on a re-projection of the three-dimensional features back to a plurality of cameras.

12. The method of claim 11 , further comprising rectifying the set of images with an original homography from a manufacture calibration, wherein the rectification is performed before the feature detection of a feature module.

13. The method of claim 11 , wherein the images of an image set comprise images each captured at a different camera comprising at least three cameras.

14. The method of claim 11 , wherein the images of the image set comprise images captured at a single camera disposed at different positions.

15. The method of claim 11 , further comprising:

discarding outliers based on feature match angle above a predetermined threshold; and

discarding outliers based on a feature match size above a predetermined threshold.

16. The method of claim 11 , further comprising discarding outliers where a feature match occurs on less than three cameras.

17. The method of claim 11 , further comprising determining a scale of the shift indicating the difference of feature coordinates between image sets divided by a displacement associated with each image captured.

18. The method of claim 11 , wherein the second homography is homography between original feature coordinates from manufacturer calibration and feature coordinates based on the re-projection of the three dimension reconstruction to one or more cameras.

19. The method of claim 18 , wherein the three dimension reconstruction estimates scene depth.

20. The method of claim 11 , wherein the method is operable ad hoc as images are captured.

21. A non-transitory, computer readable medium including code, when executed, to cause a processing device to:

detect features on a set of images;

describe features on the set of images;

determine a match between features of the image sets;

determine a shift on the matched features based on camera positions associated with the matched features;

determine a first homography between the camera positions and the determined shift;

reconstruct the detected features in three-dimensions; and

determine a second homography based on a re-projection of the three-dimensional features back to a plurality of cameras.

22. The computer readable medium of claim 21 , further comprising code, when executed, to cause the processing device to rectify the set of images with an original homography from a manufacture calibration, wherein the rectification is performed before the feature detection of a feature module.

23. The computer readable medium of claim 21 , wherein the images of an image set comprise images each captured at a different camera comprising at least three cameras.

24. The computer readable medium of claim 21 , wherein the images of the image set comprise images captured at a single camera disposed at different positions.

25. The computer readable medium of claim 21 , further comprising code, when executed, to cause the processing device to:

discard outliers based on feature match angle above a predetermined threshold; and

discard outliers based on a feature match size above a predetermined threshold.