Voxels sparse representation

Patent metadata

Field	Value
Publication number	US-10685476-B2
Application number	US-201816050953-A
Country	US
Kind code	B2
Filing date	Jul 31, 2018
Priority date	Jul 31, 2018
Publication date	Jun 16, 2020
Grant date	Jun 16, 2020

Abstract

Embodiments described herein provide an apparatus comprising a processor to project voxels from a point cloud data set into an n-DoF space, and define successively less granular supervoxels at successively higher layer of abstraction in a view of the point cloud data set, and a memory communicatively coupled to the processor. Other embodiments may be described and claimed.

First claim

What is claimed is:

1. A method, comprising:
projecting voxels from a point cloud data set into an n-Degree of Freedom (n-DoF) space, wherein n is an integer >2;
defining successively less granular supervoxels at successively higher layers of abstraction in a view of the point cloud data set;
defining a first abstraction layer;
implementing a labeling process to associate voxels in the point cloud data set into one or more clusters based upon one or more characteristics of the voxels;
assigning labels to the one or more clusters;
selecting a cluster from the one or more of clusters;
marking one or more voxels within predefined Euclidian distance of a center of the cluster;
calculating a set of expected values of color data, normal data, and coordinate data for the one or more voxels;
defining a set of supervoxels according to the set of expected values of color data, normal data, and coordinate data for the voxels;
calculating a set of covariance values of the color data, normal data, and coordinate data for the voxels; and
defining dimensions for the set of supervoxels according to the covariance values of the color data, normal data, and coordinate data for the voxels.

2. The method of claim 1 , further comprising:
combining voxels from the point cloud data set into a first set of supervoxels at a first layer of abstraction; and
combining supervoxels from the first set of supervoxels to form a second set of supervoxels at a second layer of abstraction.

3. The method of claim 1 , further comprising:
removing the labels from the one or more clusters; and
projecting the set of supervoxels into the n-DoF space.

4. The method of claim 1 , further comprising:
defining a first abstraction layer;
constructing a graph connecting the voxels in the point cloud data set within a predefined Euclidian distance, the voxels corresponding to nodes on the graph; and
setting an edge weights on the graph that are proportional to the Euclidian distance between the voxels.

5. The method of claim 4 , further comprising:
applying a small world network algorithm to combine nodes on the graph to create supervoxels;
implementing a labeling process to associate voxels in the graph set into one or more clusters based upon characteristics of the voxels; and
assigning labels to the one or more clusters.

6. The method of claim 5 , further comprising:
for each label, creating a bounding box; and
merging all voxels within the bounding box to create a super voxel.

7. The method of claim 6 , further comprising:
reconnecting the graph by increasing the Euclidian distance.

8. A non-transitory machine readable medium storing instructions which, when executed by one or more processors, cause the one or more processors to perform operations comprising: projecting voxels from a point cloud data set into an n-Degree of Freedom (n-DoF) space, wherein n is an integer >2;
defining successively less granular supervoxels at successively higher layer of abstraction in a view of the point cloud data set;
defining a first abstraction layer; implementing a labeling process to associate voxels in the point cloud data set into one or more clusters based upon one or more characteristics of the voxels;
assigning labels to the one or more clusters;
selecting a cluster from the one or more of clusters; marking one or more voxels within predefined Euclidian distance of a center of the cluster; calculating a set of expected values of color data, normal data, and coordinate data for the one or more voxels; defining a set of supervoxels according to the set of expected values of color data, normal data, and coordinate data for the voxels; calculating a set of covariance values of the color data, normal data, and coordinate data for the voxels; and defining dimensions for the set of supervoxels according to the covariance values of the color data, normal data, and coordinate data for the voxels.

9. The non-transitory machine readable medium of claim 8 , the operations additionally comprising:
combining voxels from the point cloud data set into a first set of supervoxels at a first layer of abstraction; and
combining supervoxels from the first set of supervoxels to form a second set of supervoxels at a second layer of abstraction.

10. The non-transitory machine readable medium of claim 8 , the operations additionally comprising:
removing the labels from the one or more clusters; and
projecting the set of supervoxels into the n-DoF space.

11. The non-transitory machine readable medium of claim 8 , the operations additionally comprising:
defining a first abstraction layer;
constructing a graph connecting the voxels in the point cloud data set within a predefined Euclidian distance, the voxels corresponding to nodes on the graph; and
setting an edge weights on the graph that are proportional to the Euclidian distance between the voxels.

12. The non-transitory machine readable medium of claim 11 , the operations additionally comprising:
applying a small world network algorithm to combine nodes on the graph to create supervoxels;
implementing a labeling process to associate voxels in the graph set into one or more clusters based upon characteristics of the voxels; and
assigning labels to the one or more clusters.

13. The non-transitory machine readable medium of claim 12 , the operations additionally comprising:
for each label, creating a bounding box; and
merging all voxels within the bounding box to create a super voxel.

14. The non-transitory machine readable medium of claim 13 , the operations additionally comprising:
reconnecting the graph by increasing the Euclidian distance.

15. An apparatus, comprising: a processor to project voxels from a point cloud data set into an n-Degree of Freedom (n-DoF space), wherein n is an integer >2;
define successively less granular supervoxels at successively higher layer of abstraction in a view of the point cloud data set;
define a first abstraction layer;
implement a labeling process to associate voxels in the point cloud data set into one or more clusters based upon one or more characteristics of the voxels;
assign labels to the one or more clusters;
select a cluster from the one or more of clusters; mark one or more voxels within predefined Euclidian distance of a center of the cluster; calculate a set of expected values of color data, normal data, and coordinate data for the one or more voxels; define a set of supervoxels according to the set of expected values of color data, normal data, and coordinate data for the voxels; calculate a set of covariance values of the color data, normal data, and coordinate data for the voxels; and define dimensions for the set of supervoxels according to the covariance values of the color data, normal data, and coordinate data for the voxels; and a memory communicatively coupled to the processor.

16. The apparatus of claim 15 , the processor to combine voxels from the point cloud data set into a first set of supervoxels at a first layer of abstraction, and combine supervoxels from the first set of supervoxels to form a second set of supervoxels at a second layer of abstraction.

17. The apparatus of claim 15 , the processor to remove the labels from the one or more clusters, and project the set of supervoxels into the n-DoF space.

18. The apparatus of claim 15 , the processor to define a first abstraction layer, construct a graph connecting the voxels in the point cloud data set within a predefined Euclidian distance, the voxels corresponding to nodes on the graph, and

Assignees

• Intel Corp

Inventors

• Kaufman Itay
• Distler Jonathan
• Ashkenazi Tzach
• Bortman Maria

Classifications

• G06V10/82

  using neural networks · CPC title

• G06V10/7715

  Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods · CPC title

• G06T11/26

  Drawing of charts or graphs · CPC title

• G06F18/23

  Clustering techniques · CPC title

• G06F18/24

  Classification techniques · CPC title