Camera pose estimation apparatus and method for augmented reality imaging

1. An apparatus for providing an estimate for a 3D camera pose relative to a scene from 2D image data of a 2D image frame provided by a camera, the apparatus comprising:

one or more processors; and

a memory coupled to the one or more processors and comprising computer readable instructions which, when executed by the one or more processors, cause the one or more processors to:

obtain, from a detection thread, detected 2D-3D point correspondences between a current image and a previous image of a scene,

obtain, from a tracking thread, tracked 2D-3D point correspondences between the current image and the previous image,

obtain motion model observations by comparing the current image to the previous image,

obtain edge observations in the current image and the previous image, and

determine, in a primary thread, an estimate for a 3D camera pose relative to the scene according to the detected 2D-3D point correspondences, the tracked 2D-3D point correspondences, the motion model observations, and the edge observations by minimizing a distance between edges in the current image and the previous image and edges in a 3D model of the environment projected to a screen based on a previous camera pose.

2. The apparatus according to claim 1 , wherein minimizing the distance between the edges in the current image and the previous image and the edges in the 3D model comprises numerically minimizing a 2D pixel re-projection error of the detected 2D-3D point correspondences.

3. The apparatus according to claim 1 , wherein minimizing the distance between the edges in the current image and the previous image and the edges in the 3D model comprises numerically minimizing a 2D pixel re-projection error of the tracked 2D-3D point correspondences.

4. The apparatus according to claim 1 , wherein minimizing the distance between the edges in the current image and the previous image and the edges in the 3D model comprises numerically minimizing a translation and rotation relative to the previous frame based on the notion model observations.

5. The apparatus according to claim 1 , wherein minimizing the distance between the edges in the current image and the previous image and the edges in the 3D model comprises numerically minimizing a distance between the edges in the current image and the previous image and the edges in the 3D model projected to the screen.

6. The apparatus according to claim 1 , wherein the edge observations are obtained by an observation thread, which is different from the detection thread and the tracking thread.

7. The apparatus according to claim 1 , wherein the detected 2D-3D point correspondences, the tracked 2D-3D point correspondences, the motion model observations, and the edge observations are weighted, limited and controlled against each other.

8. The apparatus according to claim 1 , wherein a Levenberg-Marquard algorithm is applied in the minimizing the distance between the edges in the current image and the previous image and the edges in the 3D model projected to the screen.

9. The apparatus according to claim 1 , wherein a robust estimator is applied in the minimizing the distance between the edges in the current image and the previous image and the edges in the 3D model projected to the screen.

10. The apparatus according to claim 9 , wherein the robust estimator comprises a Tukey estimator.

11. The apparatus according to claim 1 , wherein the 2D image data is stored in a circular memory buffer in system RAM of the apparatus.

12. The apparatus according to claim 1 , wherein the 2D image data is obtained in a separate thread than the detection thread and the tracking thread.

13. A non-transitory computer-readable medium comprising computer readable code executable by one or more processors to:

obtain, from a detection thread, detected 2D-3D point correspondences from 2D image data of a scene provided by a camera,

obtain, from a tracking thread, tracked 2D-3D point correspondences from the 2D image data,

obtain motion model observations from the 2D image data,

obtain edge observations from the 2D image data, and

determine, in a primary thread, an estimate for a 3D camera pose for the camera relative to the scene according to the detected 2D-3D point correspondences, the tracked 2D-3D point correspondences, the motion model observations, and the edge observations by minimizing a distance between edges in 2D images, obtained based on the 2D image data, and edges in a 3D model projected to a screen.

14. The non-transitory computer-readable medium of claim 13 , wherein minimizing the distance between the edges in the 2D images and the edges in the 3D model comprises numerically minimizing a 2D pixel re-projection error of the detected 2D-3D point correspondences.

15. The non-transitory computer-readable medium of claim 13 , wherein minimizing the distance between the edges in the 2D images and the edges in the 3D model comprises numerically minimizing a 2D pixel re-projection error of the tracked 2D-3D point correspondences.

16. The non-transitory computer-readable medium of claim 13 , wherein minimizing the distance between the edges in the 2D images and the edges in the 3D model comprises numerically minimizing a distance between edges in the 2D images and the edges in the 3D model projected to the screen.

17. The non-transitory computer-readable medium of claim 13 , wherein the edge observations are obtained by an observation thread, which is different from the detection and tracking threads.

18. The non-transitory computer-readable medium of claim 13 , wherein the detected 2D-3D point correspondences, the tracked 2D-3D point correspondences, the motion model observations, and the edge observations are weighted, limited and controlled against each other.

19. A method comprising:

obtaining, from a detection thread, detected 2D-3D point correspondences from 2D image data of a scene provided by a camera,

obtaining, from a tracking thread, tracked 2D-3D point correspondences from the 2D image data,

obtaining motion model observations from the 2D image data,

obtaining edge observations from the 2D image data, and

determining, in a primary thread, an estimate for a 3D camera pose for the camera relative to the scene according to the detected 2D-3D point correspondences, the tracked 2D-3D point correspondences, the motion model observations, and the edge observations by minimizing a distance between edges in 2D images, obtained based on the 2D image data, and edges in a 3D model projected to a screen.

20. The method of claim 19 , wherein the detected 2D-3D point correspondences, the tracked 2D-3D point correspondences, the motion model observations, and the edge observations are weighted, limited and controlled against each other.