Methods and apparatus for an imaging system

1. An imaging apparatus, comprising:

an image sensor comprising a pixel array configured to generate pixel data;

a storage unit coupled to the image sensor and configured to store the pixel data; and

an image processing unit communicatively coupled to the image sensor and the storage unit and configured to:

receive a first image frame of pixel data and a second image frame of pixel data, wherein the first image frame and the second image frame are captured sequentially;

detect at least two features in the first image frame and at least two features in the second image frame;

form matched features by matching the at least two features of the first image frame to the at least two features of the second image frame;

perform orientation estimation comprising:

calculating an epipole from the matched features;

estimating a rotation angle in relation to the epipole;

determining whether the matched features appear on a reference plane by calculating a 3D coordinate for each of the matched features in a reference coordinate system; and

updating the estimated rotation angle using the calculated 3D coordinate for each of the matched features and model fitting techniques until a predetermined tolerance level is reached;

validate the estimated orientation in the reference coordinate system using homography; and

estimate an average camera orientation by weighted averaging of multiple estimates of the camera orientation using quality scores as weights, and wherein a quality score measures the consistency of multiple estimates of a translation vector of the camera in the vehicle coordinate system between two image frames.

2. The imaging apparatus according to claim 1 , wherein the reference plane comprises a ground plane.

3. The imaging apparatus according to claim 1 , wherein the reference coordinate system comprises a vehicle coordinate system.

4. The imaging apparatus according to claim 1 , wherein detecting at least two features comprises:

determining an orientation of an edge incident to a detected corner and rejecting an invalid corner; and

generating a corner descriptor for the detected corner, wherein the corner descriptor comprises at least one of a position, an orientation, a subtended angle, a contrast, and an index of a neighboring corner.

5. The imaging apparatus according to claim 1 , wherein forming matched features comprises:

matching features based on the properties of neighboring features; and

generating a match descriptor for each pair of matched features, wherein the match descriptor comprises a ratio between measurements of the sizes of the matched neighborhoods.

6. The imaging apparatus according to claim 1 , wherein calculating an epipole from the matched features comprises determining a direction of movement using a change of scales between the matched features.

7. The imaging apparatus according to claim 1 , wherein the rotation angle comprises at least one of: a pitch angle, a yaw angle, and a roll angle.

8. The imaging apparatus according to claim 1 , wherein validating the estimated orientation in the reference coordinate system using homography comprises measuring the consistency of multiple estimates of a translation vector in the reference coordinate system between two image frames.

9. A method for calibrating a camera's orientation relative to a reference coordinate system comprising:

detecting at least two features on an image plane for each of a first image frame and a second image frame generated by the camera;

forming matched features, comprising matching at least two features of the first image frame to at least two features of the second image frame;

iteratively estimating a rotation angle of the camera utilizing at least two pairs of the matched features and a calculated epipole until a predetermined tolerance level is reached;

validating the estimated rotation angle in the reference coordinate system; and

estimating an average camera orientation by weighted averaging of multiple estimates of the camera orientation using quality scores as weights, and wherein a quality score measures the consistency of multiple estimates of a translation vector of the camera in the vehicle coordinate system between two image frames.

10. The method for calibrating a camera's orientation according to claim 9 , wherein the rotation angle comprises at least one of: a pitch angle, a yaw angle, and a roll angle.

11. The method for calibrating a camera's orientation according to claim 9 , wherein detecting at least two features comprises:

determining an orientation of an edge incident to a detected corner and rejecting an invalid corner; and

generating a corner descriptor for the detected corner, wherein the corner descriptor comprises at least one of a position, an orientation, a subtended angle, a contrast, and an index of a neighboring corner.

12. The method for calibrating a camera's orientation according to claim 11 , wherein the position, the orientation, and the subtended angle are rectified to account for lens distortion prior to feature matching.

13. The method for calibrating a camera's orientation according to claim 9 , wherein forming matched features comprises:

matching features based on the properties of neighboring features; and

generating a match descriptor for each pair of matched features, wherein the match descriptor comprises a ratio between measurements of the sizes of the matched neighborhoods, wherein each neighborhood is formed by at least two features.

14. The method for calibrating a camera's orientation according to claim 9 , wherein calculating an epipole from the matched features comprises determining a direction of movement of the camera using a change of scales between the matched features.

15. The method for calibrating a camera's orientation according to claim 9 , wherein estimating the camera rotation angle further comprises computing a value for a first rotation angle, and computing a remaining rotation angle from the epipole.

16. The method for calibrating a camera's orientation according to claim 15 , wherein computing a value for the first rotation angle comprises projecting 3D coordinates corresponding to at least one pair of matched features onto a 2D plane to form a set of points and fitting a line through the set of points.

17. The method for calibrating a camera's orientation according to claim 16 , further comprising adjusting at least one rotation angle according to a slope of the fitted line.

18. The method for calibrating a camera's orientation according to claim 9 , wherein validating the estimated camera rotation angle comprises measuring the consistency of multiple estimates of a translation vector in the reference coordinate system between two image frames.

19. A system for calibrating an on-vehicle camera's orientation relative to a vehicle coordinate system when the vehicle undergoes linear motion comprising:

an image sensor comprising a pixel array for capturing an image frame, wherein the pixel array comprises pixels arranged in rows and columns;

an image processing unit configured to receive sequential rows of pixel data of a

first image frame and a second image frame, comprising;

a calibration unit configured to:

detect at least two features on an image plane for each of the first image frame and the second image frame;

form matched features by matching the at least two features of the first image frame to the at least two features of the second image frame;

iteratively estimate the camera orientation utilizing the matched features and a

calculated epipole until a predetermined tolerance level is reached; and

validate the estimated camera orientation in the vehicle coordinate system; and

an output unit configured to transmit the validated camera orientation to a peripheral system and estimate an average camera orientation by weighted averaging of multiple estimates of the camera orientation using quality scores as weights, and wherein a quality score measures the consistency of multiple estimates of a translation vector of the camera in the vehicle coordinate system between two image frames.