Method and system for camera autocalibration

• US 7,356,425 B2
• Filed: 12/07/2005
• Issued: 04/08/2008
• Est. Priority Date: 03/14/2005
• Status: Expired due to Fees

• Current Assignee: UTC Fire & Security Americas Corporation, Inc. (Carrier Global Corporation)
• Original Assignee: GE Security Incorporated (Carrier Global Corporation)
• Inventors: Sebastian, Thomas Baby, Tu, Peter Henry, Krahnstoever, Nils Oliver, Rittscher, Jens, Mendonca, Paulo Ricardo
• Primary Examiner(s): Barlow; John
• Assistant Examiner(s): TEIXEIRA MOFFAT, JONATHAN CHARLES
• Application Number:
  US11/296,743
• Publication Number:
  US 20060215031A1
• Time in Patent Office: 853 Days
• Field of Search: 702/94
• US Class Current: 702/94
• CPC Class Codes: G06T 7/80 Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration

Abstract

A method for calibrating a projective camera is provided. The method includes acquiring information by detecting at least one object on a substantially flat ground plane within a field of view. A projective camera calibration is performed. A measurement uncertainty is considered to yield a plurality of camera parameters from the projective camera calibration.

Claims

1. A method for calibrating a projective camera, said method comprising:

acquiring an observation set by detecting at least one object on a substantially flat ground plane within a field of view;

determining a noise function for the observation set based on a foot-to-head homography between a two-dimensional foot location and a corresponding two-dimensional head location the foot-to-head homography analyzed to determine a homology having an eigenvector, two unit eigenvalues, and a non-unit eigenvalue; and

performing a projective camera calibration using the determined noise function.

2. A method in accordance with claim 1 wherein determining a noise function further comprises:

determining a linear solution;

determining a closest homology to the linear solution; and

refining the closest homology to determine a nonlinear refinement.

3. A method in accordance with claim 2 wherein determining a closest homology to the linear solution further comprises

decomposing the eigenvector and the eigenvalues to estimate at least one of a camera focal length, a tilt angle, a height above a ground plane and a roll angle.

4. A method in accordance with claim 3 wherein refining the closest homology further comprises refining the eigenvector and the eigenvalues to obtain a final estimate of at least one of the camera focal length, the tilt angle, the height above a ground plane and the roll angle.

5. A method in accordance with claim 4 wherein refining the eigenvector and the eigenvalues further comprises computing the final estimate using a non-linear optimization algorithm.

6. A method in accordance with claim 1 wherein determining a noise function further comprises determining a foot-to-head homology.

7. A method in accordance with claim 6 wherein determining a foot-to-head homology further comprises:

obtaining a plurality of foot locations on the ground plane;

obtaining a head location on a head plane corresponding to each foot location of the plurality of foot locations, the head location positioned with respect to the corresponding foot location along a vertical plane at a height; and

mapping an image between each foot location and the corresponding head location using the foot-to-head homology.

8. A method in accordance with claim 7 further comprising extracting the remaining camera parameters from the foot-to-head homology.

9. A method in accordance with claim 7 further comprising estimating each foot location and a corresponding head location of an image within a foreground region having a top portion, an upper-middle portion, a lower-middle portion and a bottom portion.

10. A method in accordance with claim 9 wherein estimating each foot location and a corresponding head location of an image within a foreground region further comprises:

calculating a first center of mass of the foreground region and a second order of moment of the foreground region in the upper-middle portion and the lower-middle portion;

determining an upper y-coordinate estimate of the foreground region;

determining a lower y-coordinate estimate of the foreground region;

determining a x-coordinate of a second center of mass of the top portion to provide a x-coordinate of the head location estimate; and

connecting the x-coordinate of the head location estimate with the first center of mass through the lower y-coordinate estimate to determine a x-coordinate of the foot location estimate.

11. A method in accordance with claim 10 further comprising obtaining at least one uncertainty estimate from a measurement uncertainty.

12. A method in accordance with claim 11 further comprising obtaining at least one uncertainty estimate from one of a second order of moment and a covariance matrix.

13. A method in accordance with claim 1 further comprising at least one of suppressing a false alarm during object detection, determining at least one object three-dimensional world location based on a plurality of two-dimensional measurements, locating the at least one object over time and measuring a distance between objects in the field of view.

14. A method for calibrating a projective camera, said method comprising:

acquiring image data by detecting at least one object on a ground plane within a field of view;

determining a foot-to-head homography using a vanishing point and a horizon plane; and

performing a projective camera calibration using the determined foot-to-head homography, the foot-to-head homography analyzed to determine an estimated foot-to-head homology having an eigenvector, two unit eigenvalues, and a non-unit eigenvalue.

15. A method in accordance with claim 14 wherein acquiring image data further comprises extracting a two-dimensional head location estimate and a two-dimensional foot location estimate from the image data to determine the foot-to-head homography.

16. A method in accordance with claim 14 wherein performing a projection camera calibration further comprises

decomposing the estimated foot-to-head homology to estimate at least one of a camera focal length, a tilt angle, a height above a ground plane and a roll angle.

17. A method in accordance with claim 16 further comprising refining the estimated foot-to-head homology to obtain a final estimate of at least one of the camera focal length, the tilt angle, the height above a ground plane and the roll angle.

18. A method in accordance with claim 16 wherein decomposing the estimated foot-to-head homology further comprises:

obtaining a plurality of foot locations on the ground plane;

obtaining a head location on a head plane corresponding to each foot location of the plurality of foot locations, the head location positioned with respect to the corresponding foot location along a vertical plane at a height; and

mapping an image between each foot location and the corresponding head location using the foot-to-head homology.

19. A surveillance system comprising:

an object detector configured to process a video stream and report a rectangular foreground region; and

a processor configured to;

process the reported foreground region to extract a foot location estimate on a ground plane and a corresponding head location estimate, the extracted estimates based on a foot-to-head homology having an eigenvector, two unit eigenvalues, and a non-unit eigenvalue;

perform a projective camera calibration based on at least one of the foot location estimate and the head location estimate; and

output a plurality of camera parameters based on the projective camera calibration.