System for detecting objects in streaming 3D images formed from data acquired with a medium penetrating sensor

• US 10,007,996 B2
• Filed: 03/02/2015
• Issued: 06/26/2018
• Est. Priority Date: 03/02/2015
• Status: Active Grant

• Current Assignee: Lawrence Livermore National Security LLC (Government of the United States of America)
• Original Assignee: Lawrence Livermore National Security LLC (Government of the United States of America)
• Inventors: Paglieroni, David W., Beer, N. Reginald
• Primary Examiner(s): CESE, KENNY A
• Application Number:
  US14/635,822
• Publication Number:
  US 20160260222A1
• Time in Patent Office: 1,212 Days
• CPC Class Codes: G01S 13/885 for ground probing prospecting or detecting using electromagnetic waves G01V3/12
• CPC Class Codes: G01S 7/411 Identification of targets based on measurements of radar reflectivity G01S7/415 takes precedence
• CPC Class Codes: G01V 3/12 operating with electromagnetic waves operating with millimetre waves G01V8/005
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06T 15/08 Volume rendering
• CPC Class Codes: G06T 2207/10028 Range image; Depth image; 3D point clouds
• CPC Class Codes: G06T 7/11 Region-based segmentation
• CPC Class Codes: G06T 7/194 involving foreground-background segmentation
• CPC Class Codes: G06V 20/64 Three-dimensional objects
• CPC Class Codes: H04N 13/128 Adjusting depth or disparity

Abstract

A detection system that detects subsurface objects within a medium and estimates various features of the objects is provided. The detection system receives a streaming sequence of image frames of the medium at various along-medium locations. An image frame contains voxel values (intensities) representing characteristics of the medium across the medium and in the depth (range) direction. The detection system depth-compensates the intensities for determining which voxels are part of an object using an unsupervised binary classifier. The detection system then connects object voxels into distinct objects and recursively estimates the features of those objects as the image frames stream based on the locations and intensities of the object voxels.

Claims

1. A method performed by a computing system for identifying objects within a medium, the method comprising:

accessing a sequence of image frames of the medium, each image frame corresponding to an along-medium location, each image frame corresponding to an along-medium location having, for each of a plurality of depths and for each of a plurality of cross-medium locations, a voxel at that depth and at that cross-medium location, each voxel having an intensity;

generating a depth-compensated intensity for each voxel that is a normalization of the intensities to account for variations resulting from differences in depth such that, for each depth, the depth-compensated intensity for a voxel at that depth is based on the intensity of the voxel and statistically derived characteristics of the intensities of voxels at that depth; and

analyzing the depth-compensated intensities to identify cross-medium locations of image frames with object voxels that represent part of an object.

2. The method of claim 1 wherein analysis of the depth-compensated intensities includes generating depth-integrated intensities for cross-medium locations of the image frames and applying an unsupervised binary classifier to the depth-integrated intensities of an image frame to classify whether the depth-integrated intensities across the medium represent object voxels or non-object voxels.

3. The method of claim 1 wherein analysis of the depth-compensated intensities includes applying an unsupervised binary classifier to depth-compensated intensities of an image frame to classify each voxel as being either an object voxel or a non-object voxel.

4. The method of claim 1 wherein the depth-compensated intensities are further along-medium compensated.

5. The method of claim 1 wherein a depth-compensated intensity for a depth is based on a number of standard deviations above a statistically significant intensity for that depth.

6. The method of claim 5 wherein the statistically significant intensity is based on one or more standard deviations above a mean of the intensities for that depth.

7. The method of claim 6 wherein the mean and the standard deviation for a specific depth in an image frame are based on analysis of intensities for voxels at that depth of some number of image frames prior to that image frame in the sequence.

8. The method of claim 1 wherein the image frames are generated based on return signals from a ground-penetrating radar.

9. A computing system for identifying occurrences of a physical characteristic of a medium, the computing system comprising:

a storage medium for storing images of a sequence of image frames of the medium, each image frame corresponding to an along-medium location, each image frame corresponding to an along-medium location having, for each of a plurality of depths and for each of a plurality of cross-medium locations, a voxel at that depth and at that cross-medium location, each voxel having an intensity;

a component that generates a depth-compensated intensity for each voxel that is a normalization of the intensities to account for variations resulting from differences in depth such that, for each depth, the depth-compensated intensity for a voxel at that depth is based on the intensity of the voxel and statistically derived characteristics of the intensities of voxels at that depth;

a component that analyzes the depth-compensated intensities to identify object voxels that are part of a physical characteristic of the medium; and

a component that determines which of the object voxels are part of the same occurrence of the physical characteristic.

10. The computing system of claim 9 wherein an occurrence of the physical characteristic is presence of an object within the medium.

11. The computing system of claim 9 wherein the analysis component applies an unsupervised binary classifier, with a threshold value derived from depth-compensated intensities for voxels from prior image frames, to classify whether voxels from the current image frame are object voxels or non-object voxels.

12. The computing system of claim 9 wherein the depth-compensated intensities are further along-medium compensated.

13. The computing system of claim 9 wherein the depth-compensated intensities for voxels at a specific depth are based on a number of standard deviations above a statistically significant intensity for that depth.

14. The computing system of claim 13 wherein the statistically significant intensity is based on one or more standard deviations above a mean of intensities for that depth.

15. The computing system of claim 9 wherein the object voxels are part of the same occurrence of the physical characteristic based on proximity of the object voxels to each other.

16. The computing system of claim 15 wherein the component that determines which of the identified voxels are part of the same occurrence of the physical characteristic grows occurrences by adding to an occurrence an object voxel that is connected to an object voxel within the occurrence, merging occurrences that are connected to the same object voxel, and splitting an occurrence that is connected to object voxels in the same image frame that are not connected.

17. A computer-readable storage medium storing computer-executable instructions for detecting an object in a medium, the computer-executable instruction comprising instructions that:

access a sequence of image frames of the medium, each image frame corresponding to an along-medium location, each image frame corresponding to an along-medium location having, for each of a plurality of depths and for each of a plurality of cross-medium locations, a voxel at that depth and at that cross-medium location having voxels representing various depths and cross-medium locations at that along-medium location, each voxel having an intensity;

generate a depth-compensated intensity for each voxel that is a normalization of the intensities to account for variations resulting from differences in depth such that, for each depth, the depth-compensated intensity for a voxel at that depth is based on the intensity of the voxel and statistically derived characteristics of the intensities of voxels at that depth;

analyze the depth-compensated intensities to identify object voxels that are part of an object within the medium; and

determine which of the identified voxels are part of the same object based on proximity of the voxels within the medium.

18. The computer-readable storage medium of claim 17 wherein the instructions that perform analysis further comprise instructions that apply an unsupervised binary classifier with a threshold value derived from depth-compensated intensities for voxels from prior image frames to classify each voxel from the current image frame as an object voxel or a non-object voxel.

19. The computer-readable storage medium of claim 17 wherein the depth-compensated intensities are further along-medium compensated.

20. The computer-readable storage medium of claim 19 wherein a depth-compensated intensity for a depth is based on a number of standard deviations above a statistically significant intensity for that depth derived from analysis of depth-compensated intensities at that depth within a certain number of image frames of the sequence that are prior to a current image frame.