Distance determination from image data
A computer includes a processor and a memory storing instructions executable by the processor to receive image data from a camera, generate a depth map from the image data, detect an object in the image data, apply a bounding box circumscribing the object to the depth map, mask the depth map by setting depth values for pixels in the bounding box in the depth map to a depth value of a closest pixel in the bounding box, and determine a distance to the object based on the masked depth map. The closest pixel is closest to the camera of the pixels in the bounding box.
1. A computer comprising a processor and a memory storing instructions executable by the processor to:
receive image data from a camera;
generate a depth map from the image data;
detect an object in the image data;
apply a bounding box circumscribing the object to the depth map;
determine a depth value of a closest pixel in the bounding box from among the pixels of the depth map in the bounding box, the closest pixel being closest to the camera of the pixels in the bounding box;
mask the depth map by setting depth values for the pixels in the bounding box in the depth map to the depth value of the closest pixel in the bounding box; and
determine a distance to the object based on the masked depth map.
2. The computer of claim 1 , wherein the depth map is an ordinal depth map.
3. The computer of claim 1 , wherein the image data is monocular image data.
4. The computer of claim 1 , wherein determining the distance to the object includes executing a machine-learning program, and the masked depth map is an input to the machine-learning program.
5. The computer of claim 4 , wherein the machine-learning program is a distance regressor.
6. The computer of claim 1 , wherein
the object is a first object;
the bounding box is a first bounding box;
detecting the first object in the image data includes detecting a second object in the image data; and
the instructions further include instructions to apply a second bounding box circumscribing the second object to the depth map before masking the depth map.
7. The computer of claim 6 , wherein
the pixels are first pixels;
the closest pixel is a first closest pixel;
the second bounding box includes second pixels including a second closest pixel;
the second closest pixel is closest to the camera of the second pixels in the second bounding box;
the second bounding box overlaps the first bounding box;
the first closest pixel is closer to the camera than the second closest pixel is to the camera; and
the instructions to mask the depth map include instructions to set the depth values for the second pixels in the second bounding box that are inside the first bounding box to the depth value of the first closest pixel.
8. The computer of claim 7 , wherein the instructions to mask the depth map include instructions to set the depth values for the second pixels in the second bounding box that are outside the first bounding box to the depth value of the second closest pixel.
9. The computer of claim 7 , wherein the second closest pixel is inside the first bounding box.
10. The computer of claim 1 , wherein the instructions further include instructions to determine a type of the object from the image data.
11. The computer of claim 10 , wherein determining the type of the object includes executing a machine-learning program.
12. The computer of claim 11 , wherein the image data is an input to the machine-learning program.
13. The computer of claim 11 , wherein the machine-learning program is a first machine-learning program, determining the distance to the object includes executing a second machine-learning program, and the first and second machine-learning programs are programs that have been trained together.
14. The computer of claim 10 , wherein the instructions to determine the distance to the object further include instructions to determine the distance to the object based on the type of the object.
15. The computer of claim 1 , wherein detecting the object in the image data includes executing a machine-learning program, and the image data is an input to the machine-learning program.
16. The computer of claim 15 , wherein the machine-learning program is a feature pyramid network.
17. The computer of claim 1 , wherein generating the depth map includes executing a machine-learning program, and the image data is an input to the machine-learning program.
18. The computer of claim 17 , wherein the machine-learning program is a deep neural network using an hourglass network.
19. The computer of claim 1 , wherein the instructions further include instructions to actuate a vehicle component based on the distance to the object.
20. A method comprising:
receiving image data from a camera;
generating a depth map from the image data;
detecting an object in the image data;
applying a bounding box circumscribing the object to the depth map;
determining a depth value of a closest pixel in the bounding box from among the pixels of the depth map in the bounding box, the closest pixel being closest to the camera of the pixels in the bounding box;
masking the depth map by setting depth values for the pixels in the bounding box in the depth map to the depth value of the closest pixel in the bounding box; and
determining a distance to the object based on the masked depth map.