Depth sensor data with real-time processing of scene sensor data

1. A method of improving depth sensor data in real-time based on scene sensor data, the method comprising:

estimating, with one or more processor circuits, a homography H between a depth image and a scene image of a depth-scene image pair included in a plurality of depth-scene image pairs generated by a depth-scene image sensor pair, the depth-scene image sensor pair including a depth sensor and a scene sensor;

aligning, with the one or more processor circuits, depth images of the depth-scene image pairs with corresponding scene images based on the homography H, wherein for each depth-scene image pair, the depth image is warped based on the homography H such that locations of depths in the depth image are aligned with locations of corresponding pixels in the scene image; and

correcting, with the one or more processor circuits, at least some of the aligned depth images based on image data from one or more of the scene images.

2. A digital video system comprising:

a depth-scene image sensor pair configured to capture depth-scene image pairs of a scene, the depth-scene image sensor pair including a depth sensor and a scene sensor;

means for estimating a homography H between a depth image and a scene image of a depth-scene image pair included in the depth-scene image pairs generated by the depth-scene image sensor pair;

means for aligning depth images of the depth-scene image pairs with corresponding scene images based on the homography H, wherein for each depth-scene image pair, the depth image is warped based on the homography H such that locations of depths in the depth image are aligned with locations of corresponding pixels in the scene image; and

means for correcting at least some of the aligned depth images based on image data from one or more of the scene images, wherein the means for correcting at least some of the aligned depth images comprises:

means for detecting at least one error in a region of an aligned depth image of a first depth-scene image pair based on content of a corresponding region of a scene image of the first depth-scene image pair; and

means for correcting the detected at least one error in the region of the aligned depth image, wherein the means for detecting at least one error comprises:

means for determining that at least part of an object in the scene image of the first depth-scene image pair has moved outside a range of the depth sensor;

means for increasing a cycle count in a cycle map by one for each pixel of the object outside the range if the object has moved further away, wherein the cycle map includes a cycle count for each location of the depth image of the first depth-scene image pair; and

means for decreasing the cycle count by one for each pixel of the object outside the range if the object has moved closer; and

the means for correcting the detected at least one error comprises means for modifying depths in the aligned depth image of the first depth-scene image pair based on corresponding cycle counts in the cycle map and the range of the depth sensor.

3. A non-transitory computer readable medium comprising instructions that, when executed by at least one processor in a digital video system, cause the processor to:

estimate a homography H between a depth image and a scene image of a depth-scene image pair included in a plurality of depth-scene image pairs generated by a depth-scene image sensor pair, the depth-scene image sensor pair including a depth sensor and a scene sensor;

align depth images of depth-scene image pairs generated by a depth-scene image sensor pair with corresponding scene images wherein, for each depth-scene image pair, the depth image is warped such that locations of depths in the depth image are aligned with locations of corresponding pixels in the scene image; and

correct at least some of the aligned depth images based on image data from one or more of the scene images.

4. The method of claim 1 , wherein aligning depth images further comprises:

applying an inverse of the homography H to the depth image to generate a warped depth image, wherein a depth sensor projection matrix is used to generate the warped depth image; and

applying a scene sensor projection matrix to the warped depth image to generate an aligned depth image,

wherein estimating the homography H includes estimating the homography H based on at least eight corresponding features between the depth image and the scene image of a first depth-scene image pair, wherein the homography H relates pixel coordinates of a scene image of the first depth-scene pair to point coordinates of a depth image of the first depth-scene pair.

5. The method of claim 1 , wherein the depth sensor is a time-of-flight (TOF) sensor.

6. The method of claim 1 , wherein the scene sensor is a red-green-blue (RGB) sensor.

7. The method of claim 1 , wherein correcting at least some of the aligned depth images comprises:

detecting at least one error in a region of an aligned depth image of a first depth-scene image pair based on content of a corresponding region of a scene image of the first depth-scene image pair; and

correcting the detected at least one error in the region of the aligned depth image.

8. The method of claim 7 , wherein

detecting at least one error comprises identifying a smooth region in the scene image of the first depth-scene image pair and determining that a corresponding region in the aligned depth image of the first depth-scene image pair is not smooth; and

correcting the detected at least one error comprises applying smoothing to the corresponding region.

9. The method of claim 7 , wherein

detecting at least one error comprises:

determining that at least part of an object in the scene image of the first depth-scene image pair has moved outside a range of the depth sensor;

increasing a cycle count in a cycle map by one for each pixel of the object outside the range if the object has moved further away, wherein the cycle map includes a cycle count for each location of the depth image of the first depth-scene image pair; and

decreasing the cycle count by one for each pixel of the object outside the range if the object has moved closer; and

correcting the detected at least one error comprises modifying depths in the aligned depth image of the first depth-scene image pair based on corresponding cycle counts in the cycle map and the range of the depth sensor.

10. The non-transitory computer readable medium of claim 3 , wherein the instructions that cause the processor to align depth images include instructions that cause the processor to:

apply an inverse of the homography H to the depth image to generate a warped depth image, wherein a depth sensor projection matrix is used to generate the warped depth image; and

apply a scene sensor projection matrix P to the warped depth image to generate an aligned depth image,

wherein the instructions that cause the processor to estimate the homography H include instructions that cause the processor to estimate the homography H based on at least eight corresponding features between the depth image and the scene image of a first depth-scene image pair, wherein the homography H relates pixel coordinates of a scene image of the first depth-scene pair to point coordinates of a depth image of the first depth-scene pair.

11. The non-transitory computer readable medium of claim 3 , wherein the depth sensor is a time-of-flight (TOF) sensor and the scene sensor is a red-green-blue (RGB) sensor.

12. The non-transitory computer readable medium of claim 3 , wherein the instructions that cause the processor to correct at least some of the aligned depth images include instructions that cause the processor to:

detect at least one error in a region of an aligned depth image of a first depth-scene image pair based on content of a corresponding region of a scene image of the first depth-scene image pair; and

correct the detected at least one error in the region of the aligned depth image.

13. The non-transitory computer readable medium of claim 12 , wherein the instructions that cause the processor to:

detect at least one error include instructions that cause the processor to identify a smooth region in the scene image of the first depth-scene image pair and determine that a corresponding region in the aligned depth image of the first depth-scene image pair is not smooth; and

correct the detected at least one error include instructions that cause the processor to smooth to the corresponding region.

14. The non-transitory computer readable medium of claim 12 , wherein the instructions that cause the processor to:

detect at least one error include instructions that cause the processor to:

determine that at least part of an object in the scene image of the first depth-scene image pair has moved outside a range of the depth sensor;

increase a cycle count in a cycle map by one for each pixel of the object outside the range if the object has moved further away, wherein the cycle map includes a cycle count for each location of the depth image of the first depth-scene image pair; and

decrease the cycle count by one for each pixel of the object outside the range if the object has moved closer; and

correct the detected at least one error include instructions that cause the processor to modify depths in the aligned depth image of the first depth-scene image pair based on corresponding cycle counts in the cycle map and the range of the depth sensor.