METHOD AND SYSTEM FOR FAST DENSE STEREOSCOPIC RANGING

• US 20100111444A1
• Filed: 04/24/2008
• Published: 05/06/2010
• Est. Priority Date: 04/24/2007
• Status: Active Grant

• Current Assignee: 21CT Incorporated
• Original Assignee: 21CT Incorporated
• Inventors: Coffman, Thayne R.
• Granted Patent: US 8,467,628 B2
• Time in Patent Office: Days
• US Class Current: 382/285
• CPC Class Codes: G06T 7/593 from stereo images

Abstract

A stochastic method and system for fast stereoscopic ranging includes selecting a pair of images for stereo processing, in which the pair of images are a frame pair and one of the image is a reference frame, seeding estimated values for a range metric at each pixel of the reference frame, initializing one or more search stage constraints, stochastically computing local influence for each valid pixel in the reference frame, aggregating local influences for each valid pixel in the reference frame, refining the estimated values for the range metric at each valid pixel in the reference frame based on the aggregated local influence, and post-processing range metric data. A valid pixel is a pixel in the reference frame that has a corresponding pixel in the other frame of the frame pair. The method repeats n iterations of the stochastically computing through the post-processing.

Claims

1. A stochastic method for fast stereoscopic ranging comprising:

(a) selecting a pair of images for stereo processing, wherein the pair of images are a frame pair and one of the image is a reference frame;

(b) seeding estimated values for a range metric at each pixel of the reference frame;

(c) initializing one or more search stage constraints;

(d) stochastically computing local influence for each valid pixel in the reference frame, wherein a valid pixel is a pixel in the reference frame that has a corresponding pixel in the other frame of the frame pair;

(e) aggregating local influences for each valid pixel in the reference frame;

(f) refining the estimated values for the range metric at each valid pixel in the reference frame based on the aggregated local influence; and

(g) post-processing range metric data; and

(h) repeating n iterations of (d) through (g), wherein n is a value selected in (c).

2. The method of claim 1 further comprising repeating (c) through (h) one or more times.

3. The method of claim 2 further comprising visually displaying the estimated depth values.

4. The method of claim 1 further comprising pre-processing image data from the frame pair.

5. The method of claim 4 in which pre-processing image data from the frame pair includes rectifying the reference frame.

6. The method of claim 5 in which the rectifying eliminates the relative rotations between camera positions at which each frame of the frame pair were taken.

7. The method of claim 4 in which pre-processing image data from the frame pair includes computing unit disparity vectors for each pixel in the reference frame of the frame pair.

8. The method of claim 4 in which pre-processing image data from the frame pair includes constructing masks to identify invalid pixels in the reference frame.

9. The method of claim 1 in which post-processing image data includes smoothing depth estimates.

10. The method of claim 1 in which the range metric is an estimated floating point depth value and seeding estimated values seeds estimated floating point depth values.

11. The method of claim 10 in which the estimated floating point depth value is based on a previously estimated depth value.

12. The method of claim 10 in which the outputted ranges are the range metric output from the last iteration of the search stage.

13. The method of claim 1 in which the range metric is an estimated floating point disparity value and seeding estimated values seeds estimated floating point disparity values.

14. The method of claim 13 in which the estimated floating point disparity value is based on a previously estimated disparity value.

15. The method of claim 14 in which the outputted ranges are converted from the range metric output from the last iteration of the search stage.

16. The method of claim 1 in which selecting a pair of images selects two, non-adjacent image frames from a video stream.

17. The method of claim 1 in which computing local influence for each valid pixel in the reference frame comprises:

computing match quality at the valid pixel;

perturbing a depth estimate at the valid pixel with additive noise;

computing an induced local match quality, wherein the computing applies a match quality function to the corresponding pixels determined by the perturbed depth estimate; and

computing resulting per pixel influences with the original and induced local match quality.

18. A method for initializing depth estimates in an iterative process for stereoscopic ranging, the method comprising:

projecting a first depth estimate for each one of a plurality of valid pixels in an image, the image having an implicit camera origin;

generating a 3D representation of the plurality of valid pixels using the first depth estimate;

relocating the implicit camera origin;

reprojecting the 3D representation of the plurality of valid pixels to the relocated implicit camera origin;

using Z-buffering to render a second depth estimate for each one of the plurality of valid pixels according to the relocated implicit camera origin; and

using the second depth estimate for each of the plurality of valid pixels as an initial value in a subsequent iteration of a process for stereoscopic ranging.

19. The method of claim 18 in which the first 3D representation is a point cloud.

20. A computer readable medium comprising instructions for performing a stochastic method for fast stereoscopic ranging by:

(a) selecting a pair of images for stereo processing, wherein the pair of images are a frame pair and one of the image is a reference frame;

(b) seeding estimated values for a range metric at each pixel of the reference frame;

(c) initializing one or more search stage constraints;

(d) stochastically computing local influence for each valid pixel in the reference frame, wherein a valid pixel is a pixel in the reference frame that has a corresponding pixel in the other frame of the frame pair;

(e) aggregating local influences for each valid pixel in the reference frame;

(f) refining the estimated values for the range metric at each valid pixel in the reference frame based on the aggregated local influence;

(g) post-processing range metric data; and

(h) repeating (d) through (g) for n iterations, wherein n is a value selected in (c).

21. A system for stochastic fast stereoscopic ranging comprising:

a visual sensor capable of taking images;

a computing device coupled to the visual sensor, comprising a memory, wherein the memory includes instructions for performing a stochastic method for fast stereoscopic ranging by;

(a) selecting a pair of images for stereo processing, wherein the pair of images are a frame pair and one of the image is a reference frame;

(b) seeding estimated values for a range metric at each pixel of the reference frame;

(c) initializing one or more search stage constraints;

(d) stochastically computing local influence for each valid pixel in the reference frame, wherein a valid pixel is a pixel in the reference frame that has a corresponding pixel in the other frame of the frame pair;

(e) aggregating local influences for each valid pixel in the reference frame;

(f) refining the estimated values for the range metric at each valid pixel in the reference frame based on the aggregated local influence;

(g) post-processing range metric data; and

(h) repeating (d) through (g) for n iterations, wherein n is a value selected in (c).

22. The system of claim 21 in which the computing device is a server.

23. The system of claim 22 in which the network is a wireless network.