Distance estimation using machine learning

1. A method comprising:

receiving a captured image depicting image content including an object, the captured image being captured by an image sensor located at a sensor position;

generating, using a trained first machine learning logic, a lighting-corrected image from an imitative simulation image depicting at least a portion of the image content of the captured image in a simulation style associated with an environment simulator;

generating, using a trained second machine learning logic, a depth estimation image from the lighting-corrected image, the depth estimation image indicating a relative distance between the object depicted in the captured image and the sensor position of the image sensor; and

determining an object position of the object depicted in the captured image based on the depth estimation image.

2. The method of claim 1 , wherein the image content of the captured image depicts a road segment and the object depicted in the captured image is a roadway object associated with the road segment.

3. The method of claim 1 , wherein generating the lighting-corrected image from the imitative simulation image includes:

detecting, using the trained first machine learning logic, one or more of a shadow region and a glare region in the imitative simulation image;

responsive to detecting the shadow region in the imitative simulation image, adjusting, using the trained first machine learning logic, the shadow region to reduce or remove a shadow in the shadow region of the imitative simulation image; and

responsive to detecting the glare region in the imitative simulation image, adjusting, using the trained first machine learning logic, the glare region to reduce or remove a glare in the glare region of the imitative simulation image.

4. The method of claim 1 , further comprising:

generating, using a trained third machine learning logic, the imitative simulation image from the captured image.

5. The method of claim 4 , wherein

the image content of the captured image depicts a real environment including the object;

the simulation style associated with the environment simulator includes a simulated color and a simulated texture generated by the environment simulator; and

generating the imitative simulation image from the captured image includes adjusting, using the trained third machine learning logic, one or more of a captured color of the object and a captured texture of the object to one or more of the simulated color and the simulated texture.

6. The method of claim 1 , wherein generating the depth estimation image from the lighting-corrected image includes:

mapping, using the trained second machine learning logic, one or more image pixels in the lighting-corrected image to one or more grayscale pixels in the depth estimation image.

7. The method of claim 1 , wherein

the trained first machine learning logic is a conditional-Generative Adversarial Network (GAN) and the trained second machine learning logic is another conditional-GAN; and the method includes

generating, using a trained third machine learning logic, the imitative simulation image from the captured image, the trained third machine learning logic being a cycle-GAN.

8. A system comprising:

one or more processors;

an environment simulator executable by the one or more processors to generate one or more simulation images;

a first machine learning logic executable by the one or more processors to receive a captured image depicting image content including an object, the captured image being captured by an image sensor located at a sensor position, and generate an imitative simulation image from the captured image, the imitative simulation image depicting at least a portion of the image content of the captured image in a simulation style associated with the environment simulator;

a second machine learning logic executable by the one or more processors to receive the imitative simulation image from the first machine learning logic, and generate a lighting-corrected image from the imitative simulation image;

a third machine learning logic executable by the one or more processors to receive the lighting-corrected image from the second machine learning logic, and generate a depth estimation image from the lighting-corrected image, the depth estimation image indicating a relative distance between the object depicted in the captured image and the sensor position of the image sensor; and

an object position calculator executable by the one or more processors to determine an object position of the object depicted in the captured image based on the depth estimation image.

9. The system of claim 8 , wherein the image content of the captured image depicts a road segment and the object depicted in the captured image is a roadway object associated with the road segment.

10. The system of claim 8 , wherein to generate the lighting-corrected image from the imitative simulation image, the second machine learning logic performs one or more operations including:

detecting one or more of a shadow region and a glare region in the imitative simulation image;

responsive to detecting the shadow region in the imitative simulation image, adjusting the shadow region to reduce or remove a shadow in the shadow region of the imitative simulation image; and

responsive to detecting the glare region in the imitative simulation image, adjusting the glare region to reduce or remove a glare in the glare region of the imitative simulation image.

11. The system of claim 8 , wherein

the image content of the captured image depicts a real environment including the object;

the simulation style associated with the environment simulator includes a simulated color and a simulated texture generated by the environment simulator; and

to generate the imitative simulation image from the captured image, the first machine learning logic performs one or more operations including adjusting one or more of a captured color of the object and a captured texture of the object to one or more of the simulated color and the simulated texture.

12. The system of claim 8 , wherein to generate the depth estimation image from the lighting-corrected image, the third machine learning logic performs one or more operations including:

mapping one or more image pixels in the lighting-corrected image to one or more grayscale pixels in the depth estimation image.

13. The system of claim 8 , wherein

the first machine learning logic is a cycle-Generative Adversarial Network (GAN); and

the second machine learning logic is a conditional-GAN and the third machine learning logic is another conditional-GAN.