System for mixing or compositing in real-time, computer generated 3D objects and a video feed from a film camera

1. A markerless system for mixing or compositing in real-time, computer generated 3D objects and a video feed from a video camera, to generate real-time augmented reality video, wherein the system includes augmented reality glasses, and wherein the augmented reality glasses include the video camera in which:

(a) the body of the video camera can be moved in 3D;

(b) sensors, including non-optical sensors such as an accelerometer, and/or gyroscope and/or magnetometer in or attached directly or indirectly to the video camera provide real-time positioning data defining the 3D position and 3D orientation of the video camera, or enabling the 3D position and 3D orientation of the video camera to be calculated;

(c) a pair of stereoscopic witness cameras are fixed directly or indirectly to the video camera, the pair of stereoscopic witness cameras being wide-angle stereoscopic witness cameras;

(d) that real-time positioning data, and a video flow from each stereo witness camera, is then automatically used by the system to create, recall, render or modify computer generated 3D objects; and

(e) the resulting computer generated 3D objects are then mixed in or composited with the video feed from the video camera to provide augmented reality video to the augmented reality glasses.

2. The system of claim 1 in which the sensors include a 3D range sensor.

3. The system of claim 2 in which the depth of edges is refined by reprojecting 3D range sensor depths onto a high resolution video feed of the video camera.

4. The system of claim 1 in which the pair of stereoscopic witness cameras each provides at least 100 fps, to enable the system to be fully initialized without a separate stage of purely surveying the scene to be tracked (called ‘instant survey’), but instead surveying takes place continuously whilst the camera is being used to capture video.

5. The system of claim 4 in which the pair of stereoscopic witness cameras form a stereoscopic system that enables software to process the images and, even with the camera system not being moved at all, to generate an instant 3D point cloud.

6. The system of claim 4 in which the depth of each pixel in the 3D point cloud is obtained using corresponding 2D texture patches obtained from each stereoscopic witness camera and an epi-polar line search algorithm.

7. The system of claim 4 providing a camera tracking system that combines instant survey with tracking of the film camera as the user tracks, pans, tilts the camera tracking system attached to the film camera.

8. The system of claim 1 which runs a fusion algorithm that combines optical flow data from the witness camera system with the real-time positioning data from the hardware sensors.

9. The system of claim 8 in which the fusion algorithm is based on an Extended Kalman Filter prediction/correction technique to integrate outputs from, and to re-calibrate, all the sensors, to determine the position and orientation of the camera.

10. The system of claim 9 in which the Extended Kalman Filter fusion algorithm uses confidence level data, associated with the output from each sensor, when determining how to merge the data from each sensor.

11. The system of claim 1 in which keyframes generated by the witness cameras are part of the visual tracking process and are real-time images computed at 4 different resolution levels of the witness camera video feed.

12. The system of claim 1 in which a 3D range sensor is used to enhance the accuracy of a depth measurement associated with a reconstructed 3D point, or to reject that reconstructed 3D point.

13. The system of claim 12 in which the 3D range sensor is used for real-time depth keying to enable dynamic occlusion and suppress the eventual use of a green stage.

14. The system of claim 1 that enables real-time, continuous surveying of a scene to generate a more complete point cloud defining the scene.

15. The system of claim 1 that uses a constant velocity model associated with the information provided by the sensors to predict the next position of the video camera using the previously correctly computed or confirmed position.

16. The system of claim 15 that uses that prediction to re-project a 3D point cloud onto the current frame, to enable a point matching algorithm to match points identified in the real-time video feed from the witness camera system and the projected points in the new, 3D point cloud.

17. The system of claim 16 that uses a Levenberg-Marquardt minimization scheme for camera tracking to minimize the error between the points identified in the real-time video feed from the witness camera system and the projected points in the new, 3D point cloud.

18. The system of claim 1 in which a user can use a 3D point cloud generated by the camera tracking system to define 3D masks.

19. The system of claim 1 in which the video camera and a witness camera are calibrated for frame acquisition delay using a modulated light source, such as by comparing light intensity curves associated with a flashing LED.

20. The system of claim 1 , in which

(f) the 3D position and orientation of the video camera is determined with reference to a 3D map of the real-world generated, at least in part, by using real-time 3D positioning data from the non-optical sensors plus an optical flow in which the stereo witness cameras survey a scene and software running on a processor detects natural markers in the scene that have not been manually or artificially added to that scene.

21. The system of claim 20 in which software attempts to generate a uniformly distributed point cloud in the 3D map to greatly reduce tracking losses and increase tracking accuracy, in which more parallax is generated, so the estimated camera position is more accurate.

22. The system of claim 20 that attaches rotation invariant descriptors to feature points detected in the scene to facilitate recovery of tracking.

23. A markerless method for mixing or compositing real-time, computer generated 3D objects and a video feed from a video camera, to generate augmented reality video, wherein the method uses augmented reality glasses, and wherein the augmented reality glasses include the video camera, in which:

(a) the body of the film camera can be moved in 3D;

(b) sensors, including non-optical sensors such as an accelerometer, and/or gyroscope and/or magnetometer in or attached directly or indirectly to the video camera provide real-time positioning data defining the 3D position and 3D orientation of the video camera, or enabling the 3D position and 3D orientation of the video camera to be calculated;

(c) a pair of stereoscopic witness cameras are fixed directly or indirectly to the video camera, the pair of stereoscopic witness cameras being wide-angle stereoscopic witness cameras;

(d) that real-time positioning data, and a video flow from each stereo witness camera, is then automatically used by the system to create, recall, render or modify computer generated 3D objects; and

(e) the resulting computer generated 3D objects are then mixed in or composited with the video feed from the video camera to provide augmented reality video to the augmented reality glasses.

24. The method as defined in claim 23 in which the real-time positioning data is recorded and time-code stamped to provide match-move data for post-production processes.

25. The method as defined in claim 23 to enable one or more of the following:

Using real-time data for 2D post-production

Using real-time data for post-conversion for 3D stereoscopic content

Using real-time data for native 3D stereoscopic content

3D graphics insertion

3D graphics insertion for sponsored images

3D graphics insertion that is viewer-location specific

3D graphics insertion that is viewer-specific

3D graphics insertion that is time-specific

3D graphics insertion of educational content to assist learning, in museums and interpretation centres in cultural, historic or natural sites

Measurement of the absolute or relative size of objects in the scene.

26. The method of claim 23 , in which

(f) the 3D position and orientation of the video camera is determined with reference to a 3D map of the real-world generated, at least in part, by using real-time 3D positioning data from the non-optical sensors plus an optical flow in which the stereo witness cameras survey a scene and software running on a processor detects natural markers in the scene that have not been manually or artificially added to that scene.