Apparatus for identifying objects outside of a line-of-sight

• US 10,317,515 B2
• Filed: 09/01/2017
• Issued: 06/11/2019
• Est. Priority Date: 09/01/2017
• Status: Active Grant

• Current Assignee: Wisconsin Alumni Research Foundation (University of Wisconsin System)
• Original Assignee: Wisconsin Alumni Research Foundation (University of Wisconsin System)
• Inventors: Tan, Yixuan, Wang, Xingze, Yu, Zongfu
• Primary Examiner(s): Lee, John R
• Application Number:
  US15/694,119
• Publication Number:
  US 20190072655A1
• Time in Patent Office: 648 Days
• CPC Class Codes: G01S 17/89 for mapping or imaging
• CPC Class Codes: G01S 7/4802 using analysis of echo signal for target characterisation; Target signature; Target cross-section
• CPC Class Codes: G01S 7/4914 of detector arrays, e.g. charge-transfer gates
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: G06N 3/045 Combinations of networks
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

An optical field sensor is used to make phase measurements over an area of light reflected from an object outside a field of view of the light field sensor. These phase measurements are applied to a machine learning system trained with similar phase measurements from objects outside of a field of view to identify the object within a class of objects subject to the training.

Claims

1. An optical system for non-line-of-sight object identification comprising:

a coherent light source;

an optical field sampler providing a two-dimensional array of light sensor elements, the array positionable to receive reflections of light from a non-line-of-sight object illuminated by the coherent light source to produce a corresponding array of light sensor signals; and

a machine learning system receiving the array of light sensor signals and processing them according to training by a training set of light sensor signals to identify a given object within an object class of different objects, wherein the training set links different objects of the object class to light sensor signals associated with the different objects when those different objects are not within a line-of-sight of a light sensor collecting the light sensor signals and are illuminated by a coherent light source.

2. The optical system of claim 1 wherein the machine learning system uses a feature set consisting of light phase at a set of positions over the two-dimensional array of light sensors.

3. The optical system of claim 2 wherein the light sensor elements each sense a phase of received light.

4. The optical system of claim 3 wherein the light sensor is a wavefront sensor.

5. The optical system of claim 1 wherein the light sensor elements each sense an intensity of received light.

6. The optical system of claim 1 wherein the light sensor signals of the training set provide values of reflected light of a non-line-of-sight object of the object class indirectly illuminated with coherent radiation communicated between the different objects and a multi-pixel sensor through at least one non-specular reflection.

7. The optical system of claim 1 wherein the coherent light source is selected from the group consisting of a laser and a beam expander and a coherent diode.

8. The optical system of claim 1 wherein the coherent light is un-collimated.

9. The optical system of claim 1 wherein the coherent light source provides light in a visible range.

10. The optical system of claim 1 wherein the coherent light source has a frequency linewidth of less than one gigahertz.

11. The optical system of claim 1 wherein machine learning system is a multilayer neural network.

12. The optical system of claim 11 wherein the machine learning system is a convolution neural net.

13. The optical system of claim 1 wherein the optical field sampler provides an electronic light sensor array receiving reflected light directly without intervening optics projecting a light image on the light sensor array.

14. A method of identifying a given object outside of a line-of-sight comprising:

(a) creating a training set of light sensor signals associated with different objects of an object class when those objects are not within a line-of-sight of a light sensor producing the light sensor signals and are illuminated by a coherent light source;

(b) training a machine learning system using the training set;

(c) illuminating a given object with coherent light and collecting reflected light at a location outside of the line-of-sight of the object; and

(d) processing the collected light using the trained machine learning system to identify a given object within the object class based on the collected light.

15. The method of claim 14 wherein at step (c) the object is illuminated indirectly by the coherent light after the coherent light is reflected off of at least one non-specular reflector.

16. The method of claim 14 wherein the training set created at step (a) is created in a different environment than that of the object illuminated at step (c).

17. The method of claim 14 wherein the light is collected with a spatial resolution permitting discrimination of optical speckle caused by constructive and destructive interference in reflected light from the given object.