Efficient layer-based object recognition

Patent metadata

Field	Value
Publication number	US-9355334-B1
Application number	US-201414171756-A
Country	US
Kind code	B1
Filing date	Feb 3, 2014
Priority date	Sep 6, 2013
Publication date	May 31, 2016
Grant date	May 31, 2016

Abstract

In an example embodiment, a computer-implemented method is disclosed that determines a depth image; detects an object blob in the depth image; segments the object blob into a set of layers; and compares the set of layers associated with the object blob with a set of object models to determine a match. Comparing the set of layers with the set of object models can include determining a likelihood of the object blob as belonging to each of the object models and determining the object blob to match a particular object model based on the likelihood. Further, determining the likelihood of the object blob as belonging to the one of the object models can include computing a recognition score for each layer of the set of layers; and aggregating the recognition score of each layer of the set of layers.

First claim

What is claimed is:

1. A computer-implemented method comprising:
determining, using one or more computing devices, a depth image;
detecting, using the one or more computing devices, an object blob in the depth image;
segmenting, using the one or more computing devices, the object blob into a set of layers;
extracting, using the one or more computing devices, one or more geometric properties for each layer of the set of layers by fitting a line and a parabola to data points of each layer of the set of layers segmented from the object blob; and
comparing, using the one or more computing devices, the one or more geometric properties associated with each layer of the set of layers with a set of object models to determine a match.

2. The computer-implemented method of claim 1 , wherein comparing the one or more geometric properties associated with each layer of the set of layers with the set of object models includes:
determining, using the one or more computing devices, a likelihood of the object blob as belonging to each of the object models;
determining the object blob to match a particular object model based on the likelihood.

3. The computer-implemented method of claim 2 , wherein determining the likelihood of the object blob as belonging to the one of the object models includes:
computing, using the one or more computing devices, a recognition score for each layer of the set of layers; and
aggregating, using the one or more computing devices, the recognition score of each layer of the set of layers.

4. The computer-implemented method of claim 1 , wherein extracting the one or more geometric properties for each layer of the set of layers includes determining, using the one or more computing devices, a curvature associated with the set of layers, and wherein comparing the one or more geometric properties associated with each layer of the set of layers with the object models includes evaluating the curvature using the object models.

5. The computer-implemented method of claim 1 , wherein the one or more geometric properties includes a multi-dimensional vector containing properties associated with layer curvature.

6. The computer-implemented method of claim 1 , wherein the one or more geometric properties reflect one or more of a size, curvature, curve fit, and shape fit.

7. The computer-implemented method of claim 1 , wherein detecting the object blob in the depth image includes:
detecting a plurality of blobs associated with a plurality of objects depicted by the depth image; and
classifying each blob from the plurality of blobs as a person blob or other blob type based on a shape of the blob.

8. The computer-implemented method of claim 7 , further comprising:
discarding each blob from the plurality of blobs that is classified as another blob type.

9. The computer-implemented method of claim 1 , wherein the object blob is a person blob and the method further comprises recognizing, using the one or more computing devices, a person associated with the person blob based on the match.

10. The computer-implemented method of claim 1 , wherein the set of layers includes one or more horizontal layers.

11. The computer-implemented method of claim 1 , wherein the sensor includes one or more of a stereo camera, a structured light camera, and a time-of-flight camera.

12. A computer program product comprising a non-transitory computer-readable medium storing a computer-readable program, wherein the computer-readable program, when executed on one or more computing devices, causes the one or more computing devices to:
determine a depth image;
detect an object blob in the depth image;
segment the object blob into a set of layers;
extract one or more geometric properties for each layer of the set of layers by fitting a line and a parabola to data points of each layer of the set of layers segmented from the object blob; and
compare the one or more geometric properties associated with each layer of the set of layers with a set of object models to determine a match.

13. The computer program product of claim 12 , wherein to compare the one or more geometric properties associated with each layer of the set of layers with the set of object models includes:
determining a likelihood of the object blob as belonging to each of the object models; and
determining the object blob to match a particular object model based on the likelihood.

14. The computer program product of claim 13 , wherein determining the likelihood of the object blob as belonging to the one of the object models includes:
computing a recognition score for each layer of the set of layers; and
aggregating the recognition score of each layer of the set of layers.

15. The computer program product of claim 12 , wherein
to extract the one or more geometric properties for each layer of the set of layers includes determining a curvature associated with the set of layers, and to compare the one or more geometric properties associated with each layer of the set of layers with the object models includes evaluating the curvature using the object models.

16. The computer program product of claim 12 , wherein the one or more geometric properties includes a multi-dimensional vector containing properties associated with layer curvature.

17. The computer program product of claim 12 , wherein the one or more geometric properties reflect one or more of a size, curvature, curve fit, and shape fit.

18. The computer program product of claim 12 , wherein to detect the object blob in the depth image includes:
detecting a plurality of blobs associated with a plurality of objects depicted by the depth image; and
classifying each blob from the plurality of blobs as a person blob or other blob type based on a shape of the blob.

19. The computer program product of claim 18 , wherein the computer-readable program, when executed on the one or more computing devices, further causes the one or more computing devices to:
discard each blob from the plurality of blobs that is classified as another blob type.

20. The computer program product of claim 12 , wherein the object blob is a person blob and the computer-readable program, when executed on the one or more computing devices, further causes the one or more computing devices to recognize a person associated with the person blob based on the match.

21. The computer program product of claim 12 , wherein the set of layers includes one or more horizontal layers.

22. The computer program product of claim 12 , wherein the sensor includes one or more of a stereo camera, a structured light camera, and a time-of-flight camera.

23. A system comprising:
one or more processors;
one or more memories storing instructions that, when executed by the one or more processors, cause the system to:
determine a depth image;
detect an object blob in the depth image;
segment the object blob into a set of layers;
extract one or more geometric properties for each layer of the set of layers by fitting a line and a parabola to data points of each layer of the set of layers segmented from the object blob; and
compare the one or more geometric properties associated with each layer of the set of layers with a set of object models to determine a match.

24. The system of claim 23 , wherein to compare the one or more geometric properties associated with each layer of the set of layers with the set of object models includes:
determining a likelihood of the object

Assignees

• Toyota Motor Co Ltd

Inventors

• Martinson Eric

Classifications

• G06T7/0079

  Physics · mapped topic

• G06K9/00362

  Physics · mapped topic

• G06K9/00214

  Physics · mapped topic

• G06K9/6202 Primary

  Physics · mapped topic

• G06V40/103 Primary

  Static body considered as a whole, e.g. static pedestrian or occupant recognition · CPC title