Online per-feature descriptor customization

Patent metadata

Field	Value
Publication number	US-9811760-B2
Application number	US-201514814630-A
Country	US
Kind code	B2
Filing date	Jul 31, 2015
Priority date	Jul 31, 2015
Publication date	Nov 7, 2017
Grant date	Nov 7, 2017

Abstract

An image processing system includes a processing device having a memory device for storing computer-executable instructions. The processing device is programmed to define a first descriptor in a first image, generate a mask from the first descriptor, and define a second descriptor in a second image. The processing device is further programmed to compare the first descriptor to the second descriptor to define a first error vector, determine a second error vector by applying a mask, and compute the total error of the second error vector to determine an error between the first descriptor and the second descriptor.

First claim

The invention claimed is:

1. An image processing system comprising:
a processor having a memory storing computer-executable instructions, wherein the processor is programmed to:
define a first descriptor in a first image;
generate a mask from the first image;
define a second descriptor in a second image, wherein the first descriptor includes a plurality of first test points and wherein the second descriptor includes a plurality of second test points;
compare the first descriptor to the second descriptor to define a first error vector;
determine a second error vector by applying the mask, wherein the mask reduces the plurality of first test points to a subset of the plurality of first test points based at least in part on a robustness to viewpoint change of each of the plurality of first test points; and
determine an error between the first descriptor and the second descriptor using the second error vector.

2. The image processing system of claim 1 , wherein the processor is programmed to blur the first image before defining the first descriptor and blur the second image before defining the second descriptor.

3. The image processing system of claim 1 , wherein the mask includes a bit value for each of the subset of the plurality of first test points, and wherein each bit value defines the robustness of each of the subject of the plurality of first test points.

4. The image processing system of claim 1 , wherein generating the mask includes:
testing the plurality of first test points for robustness to viewpoint change;
identifying certain test points of the plurality of first test points as unreliable test points as a result of testing the plurality of first test points for robustness;
suppressing the unreliable test points from the plurality of first test points.

5. The image processing system of claim 1 , wherein comparing the first descriptor to the second descriptor includes applying an exclusive OR (XOR) operation to the first descriptor and the second descriptor.

6. The image processing system of claim 1 , wherein the second error vector is determined as a result of applying an AND operation to the mask and the first error vector.

7. The image processing system of claim 6 , wherein applying the AND operation suppresses errors in the first error vector.

8. The image processing system of claim 1 , wherein comparing the first descriptor to the second descriptor includes applying a popcount (POPCNT) operation to the second error vector, and wherein the error between the first descriptor and the second descriptor is determined as a result of applying the popcount operation to the second error vector.

9. The image processing system of claim 1 , further comprising a camera configured to capture the first image and the second image, wherein the second image represents a viewpoint change of the camera relative to the first image.

10. The image processing system of claim 9 , wherein the first image is used to generate a plurality of descriptors and masks, and the second image is used to generate a plurality of descriptors, and wherein the descriptors and masks generated from the first image are compared to descriptors determined from a plurality of subsequent images.

11. A method comprising
defining a first descriptor in a first image;
generating a mask from the first descriptor;
defining a second descriptor in a second image, wherein the first descriptor includes a plurality of first test points and wherein the second descriptor includes a plurality of second test points;
comparing the first descriptor to the second descriptor to define a first error vector;
determining a second error vector by applying a mask, wherein the mask reduces the plurality of first test points to a subset of the plurality of first test points based at least in part on a robustness to viewpoint change of each of the plurality of first test points; and
determining an error between the first descriptor and the second descriptor from the second error vector.

12. The method of claim 11 , further comprising:
blurring the first image before defining the first descriptor; and
blurring the second image before defining the second descriptor.

13. The method of claim 11 , wherein generating the mask includes:
testing the plurality of first test points for robustness to viewpoint change;
identifying certain test points of the plurality of first test points as unreliable test points as a result of testing the plurality of first test points for robustness; and
filtering the unreliable test points from the plurality of first test points.

14. The method of claim 11 , wherein comparing the first descriptor to the second descriptor includes applying an exclusive OR (XOR) operation to the first descriptor and the second descriptor.

15. The method of claim 11 , wherein determining the second error vector includes determining the second error vector as a result of applying an AND operation to the mask and the first error vector to suppress errors in the first error vector.

16. The method of claim 11 , wherein comparing the first descriptor to the second descriptor includes applying a popcount (POPCNT) operation to the second error vector, and wherein the error between the first descriptor and the second descriptor is determined as a result of applying the popcount operation to the second error vector.

Assignees

• Ford Global Tech Llc
• Univ Michigan Regents

Inventors

• Richardson Andrew
• Olson Edwin

Classifications

• G06V10/776

  Validation; Performance evaluation · CPC title

• G06V10/462 Primary

  Salient features, e.g. scale invariant feature transforms [SIFT] · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06T1/00

  General purpose image data processing · CPC title

• G06T1/20 Primary

  Processor architectures; Processor configuration, e.g. pipelining · CPC title