Method for simulating the effects of the optical quality of windshield

Patent metadata

Field	Value
Publication number	US-12223584-B2
Application number	US-202118248317-A
Country	US
Kind code	B2
Filing date	Sep 28, 2021
Priority date	Oct 8, 2020
Publication date	Feb 11, 2025
Grant date	Feb 11, 2025

Abstract

A method for simulating the effects of the optical quality of windshield onto the image recording quality of a digital image recording device, in particular a digital image recording device for an advanced or automated driver-assistance system. The method is based on a combination of an adapted stochastic ray tracing method and of a convolutional image processing to simulate the effects of the optical distortions of a windshield on the image recording quality of a digital image recording device. The simulation is performed from a measured optical quality function of the windshields related to those optical distortions.

First claim

The invention claimed is:

1. A computer implemented method for generating a map of point spread function convolutional kernels which simulates effects of optical distortions of a windshield on an image recording quality of a digital image recording device,
wherein said method takes as input a measured optical quality function related to the optical distortions of the windshield,
wherein said method provides as output a map of point spread function convolutional kernels simulating the effects of optical distortions of a windshield on the image recording quality of the digital image recording device, wherein said method comprises the following steps:
(a) modelling at least one sheet of transparent mineral glass comprising two main parallel faces, wherein a surface of at least one of said two main parallel faces is textured with the measured optical quality function, and wherein the sheet of mineral glass is placed in front of a grid of point sources and is inclined, in respect to said grid, with an installation angle of said windshield in a transporting vehicle;
(b) calculating, with a stochastic ray tracing method, a global illuminance arriving through the modelled inclined sheet of transparent mineral glass from the grid used as a light source, and
(c) computing a projection of the global illuminance for each point source of the grid from the view frustum of a virtual camera with an optical camera model, wherein said virtual camera is placed in front of an opposite main face of the sheet of transparent mineral glass with respect to the grid used as a light source and at a position corresponding to an installation position of the digital image recording device, wherein the projected grid of point sources is a map of point spread function convolutional kernels, wherein each point spread function convolutional kernel is associated to a point source of the grid.

2. The computer implemented method according to claim 1 , further comprising, after step (c), a step of interpolating point spread function convolutional kernels for additional non projected point sources located between source points of the grid from the point spread function convolutional kernels associated to neighbour projected point sources of said non projected point sources.

3. The computer implemented method according to claim 2 , wherein the interpolating step is performed from only selected parameters of the point spread function convolutional kernels associated to neighbour projected point sources.

4. The computer implemented method according to claim 1 , further comprising step of computing a map of position shifts, wherein each position shifts of said map is the position shift between a point source of the grid and the corresponding point spread function convolutional kernel, wherein the map of point spread function convolutional kernels comprises said map of position shifts as output.

5. The computer implemented method according to claim 1 , wherein the optical camera model is represented as a recording sensor pixels in size and optical system, wherein the half size of the grid of point source is calculated with the following formulas

{

S
x

=

D
×
tan
⁢

(

H
×
π

1
⁢
8
⁢
0

)

S
y

=

D
×
tan
⁢

(

V
×
π

1
⁢
8
⁢
0

)

S
z

=
D

wherein

H
=

H
⁢
F
⁢
O
⁢
V

2

⁢

and
⁢

V

=

N
x

N
y

×
H

,

 wherein HFOV is the horizontal view frustum of a virtual camera and D is a distance between the sensor and the grid of point sources.

6. The computer implemented method according to claim 5 , wherein the grid comprises a point source for each pixel of the sensor or one same point source for a given number of pixels of the sensor.

7. The computer implemented method according to claim 1 , wherein steps (a)-(c) are reiterated for grid of point sources located at different distances from the virtual camera in order to form a 3D map of point source function convolutional kernels for a given region of the view frustum of said camera.

8. The computer implemented method according to claim 1 , wherein the measured optical quality function related to the optical distortions of the windshield is the measured transmitted wavefront error of the windshield, the measured surfaces profiles and/or the measured distribution of complex refractive index.

9. The computer implemented method according to claim 1 , wherein the grid is digitally preprocessed as an environment map projected onto an inside side of an environment sp

Assignees

• Saint Gobain

Inventors

• Laudereau Jean-Baptiste
• Berard Mathieu
• Severin-Fabiani Tatiana
• Mielec Nicolas

Classifications

• G06T5/80

  Geometric correction · CPC title

• G06T15/20

  Perspective computation · CPC title

• G06V20/56

  exterior to a vehicle by using sensors mounted on the vehicle · CPC title

• G06F30/27 Primary

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• G06T2207/20081

  Training; Learning · CPC title