Dark flash normal camera

Patent metadata

Field	Value
Publication number	US-12418724-B2
Application number	US-202118246609-A
Country	US
Kind code	B2
Filing date	Nov 9, 2021
Priority date	Nov 9, 2020
Publication date	Sep 16, 2025
Grant date	Sep 16, 2025

Abstract

Techniques of estimating surface normals and reflectance from poorly-lit images includes using, in addition to an RGB image of a subject of a set of subjects, an image illuminated with near-infrared (NIR) radiation to determine albedo and surface normal maps for performing an image relighting, the image being captured with the NIR radiation from essentially the same perspective from which the RGB image was captured. In some implementations, a prediction engine takes as input a single RGB image and a single NIR image and estimates surface normals and reflectance from the subject.

First claim

What is claimed is:

1. A method comprising:
receiving image training data representing a plurality of color images and a plurality of near-infrared (NIR) images, each of the plurality of color images being captured with a visible spectrum illumination source, each of the plurality of NIR images being captured with a NIR illumination source; and
generating a prediction engine based on the image training data, the prediction engine being configured to produce an estimated surface normal map of a user and an estimated reflectance map of the user from a color image of the user and a NIR image of the user, the color image of the user and the NIR image of the user being captured within a time period less than a threshold time period from perspectives different by less than a threshold perspective.

2. The method as in claim 1 , wherein the plurality of color images and the plurality of NIR images includes, respectively, subsets of the plurality of color images and subsets of the plurality of NIR images, each subset of the plurality of color images and each subset of the plurality of NIR images including an image of a respective subject of a set of subjects in a pose through a plurality of illumination conditions, and
wherein the method further comprises:
prior to generating the prediction engine, performing, for each of the set of subjects, a semantic segmentation operation on a color image of a corresponding subset of the plurality of color images to produce a label image for that subject, the label image having a specified number of classes into which each of a plurality of pixels of the color image is categorized, the label image for that subject being included in the image training data.

3. The method as in claim 2 , wherein each subset of the plurality of color images and each subset of the plurality of NIR images are captured using an image capture arrangement, the image capture arrangement including, respectively, a plurality of color sources, a plurality of NIR illumination sources, and color and NIR illumination detectors, the plurality of color sources and the plurality of NIR illumination sources being arranged in a geometrical pattern surrounding the color and NIR illumination detectors.

4. The method as in claim 3 , wherein the plurality of color sources and the plurality of NIR illumination sources are arranged in corners of a rectangle surrounding the color and NIR illumination detectors.

5. The method as in claim 3 , wherein each of the plurality of illumination conditions includes one of the plurality of color illumination sources and one of the plurality of NIR illumination sources producing illumination, and all other of the plurality of color illumination sources and all of the other of the plurality of NIR illumination sources not producing illumination.

6. The method as in claim 3 , wherein the color and NIR illumination detectors include a first NIR camera, a second NIR camera, and a color camera, the first NIR camera and the color camera being misaligned by an amount less than a specified misalignment threshold.

7. The method as in claim 3 , wherein the image capture arrangement further includes a NIR dot projector configured to project a dot speckle pattern on the subject, the dot speckle pattern being temporally interleaved with illumination emitted by an NIR illumination source of the plurality of NIR illumination sources.

8. The method as in claim 1 , wherein the color image of the user and the NIR image of the user are captured essentially simultaneously.

9. The method as in claim 1 , wherein the prediction engine includes a first branch and a second branch, the first branch configured to generate surface normal maps, the second branch configured to output predicted reflectance maps.

10. The method as in claim 9 , wherein the prediction engine includes a neural network having a unet encoder-decoder architecture with skip level connections, the unet encoder-decoder architecture including an encoder and decoder, the encoder including a set of blocks, each of the set of blocks including a set of convolution layers and a set of ReLU activation layers, the decoder being configured to output surface normal maps in the first branch and predicted reflectance maps in the second branch.

11. The method as in claim 9 , wherein generating the prediction engine includes:
supervising a training operation on the prediction engine using a stereo loss and a photometric loss.

12. The method of claim 11 , wherein generating the prediction engine further includes:
generating the photometric loss based on a rendering from the estimated surface normal map and the estimated reflectance map under an illumination condition of the plurality of illumination conditions.

13. The method as in claim 12 , wherein the estimated reflectance map includes a diffuse component and a specular component, and
wherein generating the photometric loss includes:
using a Lambertian reflectance model to generate a diffuse component of the estimated reflectance map; and
using a Blinn-Phong bidirectional reflectance distribution function (BRDF) to generate a specular component of the estimated reflectance map.

14. The method as in claim 12 , wherein generating the photometric loss includes:
generating a binary shadow map based on a stereo depth map and a position of a light source used in generating a color image of the plurality of color images;
generating an observed intensity map based on the estimated reflectance map; and
generating, as the photometric loss, a Hadamard product of the binary shadow map and a difference between the observed intensity map and the color image.

15. The method as in claim 12 , further comprising:
acquiring a stereo depth map;
performing a smoothing operation on the stereo depth map to produce a smoothed stereo depth map; and
generating a stereo loss based on the estimated surface normal map and gradients of the smoothed stereo depth map.

16. The method as in claim 15 , wherein generating the stereo loss includes:
generating, as an L1 vector loss, an L1 norm of a difference between the estimated surface normal map and the gradients of the smoothed stereo depth map;
generating, as an angular loss, an inner product of the estimated surface normal map and the gradients of the smoothed stereo depth map; and
generating, as the stereo loss, a difference between the L1 vector loss and the angular loss.

17. The method as in claim 1 , further comprising:
using the prediction engine to produce the estimated surface normal map of the user and the estimated reflectance map of the user from the color image of the user and the NIR image of the user.

18. The method as in claim 1 , wherein the color image of the user is a single color image, and the NIR image of the user is a single NIR image.

19. A computer program product comprising a non-transitory storage medium, the computer program product including code that, when executed by processing circuitry, causes the processing circuitry to perform a method, the method comprising:
receiving image training data representing a plurality of color images and a plurality of near-infrared (NIR) images, each of the plurality of color images being captured with a visible spectrum illumination source, each of the plurality of NIR images being captured with a NIR illumination source; and
generating a prediction engine based on the image training data, the prediction engine being configured to produce an estimated surface normal map of a user and an estimated reflectance map of the user fr

Assignees

• Google Llc

Inventors

• Lawrence Jason
• Achar Supreeth
• Xia Zhihao

Classifications

• H04N23/20

  for generating image signals from infrared radiation only · CPC title

• G06T5/70

  Denoising; Smoothing · CPC title

• G06T2207/10012

  Stereo images · CPC title

• H04N23/84

  for processing colour signals · CPC title

• H04N23/11 Primary

  for generating image signals from visible and infrared light wavelengths · CPC title