Systems and methods for estimating pose

• US 9,213,938 B2
• Filed: 06/22/2011
• Issued: 12/15/2015
• Est. Priority Date: 06/22/2010
• Status: Active Grant

• Current Assignee: University of Florida Research Foundation, Incorporated (State University System of Florida)
• Original Assignee: University of Florida Research Foundation, Incorporated (State University System of Florida)
• Inventors: Mehta, Siddhartha Satish, Dixon, Warren E., Barooah, Prabir
• Primary Examiner(s): Nguyen, Huy T
• Application Number:
  US13/701,153
• Publication Number:
  US 20130107039A1
• Time in Patent Office: 1,637 Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06N 5/00 Computing arrangements using knowledge-based models
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 2207/20076 Probabilistic image processing
• CPC Class Codes: G06T 7/73 using feature-based methods
• CPC Class Codes: G06V 10/10 Image acquisition document image scanning and transmission H04N1/00; control of digital cameras H04N23/60

Abstract

In one embodiment, a method for estimating pose includes generating hypotheses for a pose component under consideration, identifying a mode of a probability mass function associated with the hypotheses, extracting low-noise hypotheses, and averaging the low-noise hypotheses to obtain a pose component estimate.

Claims

1. A method for guidance and navigation of an autonomous vehicle, the autonomous vehicle comprising at least one camera, the method comprising:

processing a plurality of images of an environment of the autonomous vehicle captured using the at least one camera of the autonomous vehicle over a period of time to track one or more objects within the environment, wherein processing the plurality of images to track the one or more objects in the environment comprises processing at least two images of the plurality of images to estimate a pose of a first object in the environment, wherein processing the at least two images to estimate the pose of the first object comprises;

generating hypotheses for a pose component of pose of the first object;

identifying a mode of a probability mass function associated with the hypotheses;

extracting low-noise hypotheses; and

averaging the low-noise hypotheses to obtain an estimate for the pose component for the pose of the first object.

2. The method of claim 1, wherein generating hypotheses comprises generating rotation hypotheses for a rotation of the first object in the plurality of images.

3. The method of claim 2, further comprising converting the rotation hypotheses into unit quaternions in four-dimensional space to create a four-dimensional unit sphere.

4. The method of claim 3, further comprising dividing the four-dimensional unit sphere into multiple bins of equal size.

5. The method of claim 4, further comprising identifying the bin that contains the largest number of quaternions.

6. The method of claim 5, wherein identifying the mode comprises setting the mode as a center of the identified bin or a center of a dominant cluster of quaternions within the identified bin.

7. The method of claim 1, wherein generating hypotheses comprises generating unit translation hypotheses that create a three-dimensional unit sphere.

8. The method of claim 7, further comprising dividing the three-dimensional unit sphere into multiple bins of equal size.

9. The method of claim 8, further comprising identifying the bin that contains the largest number of hypotheses.

10. The method of claim 9, wherein identifying the mode comprises setting the mode as a center of the identified bin or a center of a dominant cluster of hypotheses within the identified bin.

11. The method of claim 1, wherein extracting low-noise hypotheses comprises identifying the hypotheses that are within a predetermined distance of the mode.

22. The method of claim 1, wherein:

hypotheses are low-noise when within a threshold amount of the mode; and

extracting the low-noise hypotheses comprises identifying hypotheses that are within the threshold amount of the mode.

23. The method of claim 1, wherein generating hypotheses for a pose component of pose of the first object comprises:

identifying matching feature points of the plurality of images; and

processing the matching feature points to determine the hypotheses.

24. The method of claim 2, wherein generating hypotheses further comprises generating unit translation hypotheses that create a three-dimensional unit sphere.

25. The method of claim 1, wherein:

the pose comprises two pose components, a first pose component being rotation and a second pose component being translation;

the method further comprises separately estimating rotation and translation;

estimating rotation comprises performing the acts of generating hypotheses, identifying the mode, extracting low-noise hypotheses, and averaging the low-noise hypotheses; and

estimating translation comprises performing the acts of generating hypotheses, identifying the mode, extracting low-noise hypotheses, and averaging the low-noise hypotheses.

12. A system for estimating pose, the system comprising:

a camera configured to capture sequential images;

a computer comprising memory that stores pose estimation algorithm configured to estimate pose of an object from the sequential images by;

generating hypotheses for a pose component under consideration,

identifying a mode of a probability mass function associated with the hypotheses,

extracting low-noise hypotheses, and

averaging the low-noise hypotheses to obtain a pose component estimate.

13. The system of claim 12, wherein generating hypotheses comprises generating rotation or translation hypotheses.

14. The system of claim 12, wherein the algorithm is further configured to create a unit sphere, each hypothesis being located on a surface of the unit sphere.

15. The system of claim 14, wherein the algorithm is further configured to divide the unit sphere into multiple bins of equal size.

16. The system of claim 15, wherein the algorithm is further configured to identify the bin that contains the largest number of hypotheses.

17. The system of claim 16, wherein the algorithm is further configured to set the mode as the center of the identified bin or the center of a dominant cluster of hypotheses within the identified bin.

18. The system of claim 12, wherein extracting low-noise hypotheses comprises identifying the hypotheses that are within a predetermined distance of the mode.

19. A non-transitory computer-readable medium having encoded thereon executable instructions that, when executed by at least one processor, cause the at least one processor to carry out a method for guidance and navigation of an autonomous vehicle, the autonomous vehicle comprising at least one camera, the method comprising:

processing a plurality of images of an environment of the autonomous vehicle captured using the at least one camera of the autonomous vehicle over a period of time to track one or more objects within the environment, wherein processing the plurality of images to track the one or more objects in the environment comprises processing at least two images of the plurality of images to estimate a pose of a first object in the environment, wherein processing the at least two images to estimate the pose of the first object comprises;

identifying matching feature points between sequential images;

generating hypotheses for a pose component of the pose of the first object;

identifying a mode of a probability mass function associated with the hypotheses;

extracting low-noise hypotheses; and

averaging the low-noise hypotheses to obtain an estimate for the pose component of the pose of the first object.

20. The computer-readable medium of claim 19, wherein the method further comprises:

creating a unit sphere, each hypothesis being located on a surface of the sphere;

dividing the unit sphere into multiple bins of equal size;

identifying the bin that contains the largest number of hypotheses; and

setting the mode as the center of the identified bin or the center of a dominant cluster of hypotheses within the identified bin.

21. The computer-readable medium of claim 19, wherein extracting low-noise hypotheses comprises identifying the hypotheses that are within a predetermined distance of the mode.

26. An autonomous vehicle comprising:

at least one camera;

at least one processor; and

at least one storage medium having encoded thereon executable instructions that, when executed by the at least one processor, cause the at least one processor to carry out a method for guidance and navigation of the autonomous vehicle, the method comprising;

processing a plurality of images captured by the at least one camera over a period of time to track a first object in an environment of the autonomous vehicle, wherein tracking the first object comprises estimating a relative pose of the first object between a first image and a second image of the plurality of images, wherein estimating the relative pose of the first object comprises;

estimating a translation of the first object between the first image and the second image based at least in part on the first image and the second image, and

separately from estimating the translation, estimating a rotation of the first object between the first image and the second image based at least in part on the first image and the second image.

27. The autonomous vehicle of claim 26, wherein estimating the rotation comprises:

generating hypotheses for the rotation, each of the hypotheses identifying an estimated rotation;

identifying, from the hypotheses, low-noise hypotheses at least in part by identifying a mode of the estimated rotations and a subset of the hypotheses within a threshold amount of the mode, the subset of the hypotheses being the low-noise hypotheses; and

averaging estimated rotations of the low-noise hypotheses to obtain the rotation for the first object.

28. The autonomous vehicle of claim 26, wherein estimating the translation comprises:

generating hypotheses for the translation, each of the hypotheses identifying an estimated translation;

identifying, from the hypotheses, low-noise hypotheses at least in part by identifying a mode of the estimated translations and a subset of the hypotheses within a threshold amount of the mode, the subset of the hypotheses being the low-noise hypotheses; and

averaging estimated translations of the low-noise hypotheses to obtain the translation for the first object.