Device and a method for assigning labels of a plurality of predetermined classes to pixels of an image

1. A device ( 12 ) for assigning a label of one of a plurality of predetermined classes to each pixel of an image ( 13 ), the device ( 12 ) is configured to

receive an image ( 13 ) captured by a camera ( 11 ), the image ( 13 ) comprising a plurality of pixels;

use an encoder convolutional neural network to generate probability values for each pixel, each probability value indicating the probability that the respective pixel is associated with one of the plurality of predetermined classes;

generate for each pixel a class prediction value from the probability values, the class prediction value predicting the class of the plurality of predetermined classes the respective pixel is associated with;

use an edge detection algorithm to predict boundaries between objects shown in the image ( 13 ), the class prediction values of the pixels being used as input values of the edge detection algorithm; and

assign a label of one of the plurality of predetermined classes to each pixel of the image ( 13 ) based on the predicted boundaries.

2. The device ( 12 ) as claimed in claim 1 , wherein the device ( 12 ) is further configured to:

create a mask ( 30 ) that covers pixels of the predicted boundaries; and

filter the pixels of the predicted boundaries covered by the mask ( 30 ).

3. The device ( 12 ) as claimed in claim 2 , wherein the device ( 12 ) is further configured to add pixels in local neighborhoods of the predicted boundaries to the mask ( 30 ).

4. The device ( 12 ) as claimed in claim 1 , wherein the device ( 12 ) is further configured to generate for each pixel the class predicting value by selecting the class associated with the probability value having a highest probability value for the respective pixel.

5. The device ( 12 ) as claimed in claim 1 , wherein the camera ( 11 ) is mounted on a vehicle.

6. The device ( 12 ) as claimed in claim 1 , wherein the edge detection algorithm is a Sobel edge detection algorithm.

7. The device ( 12 ) as claimed in claim 1 , wherein kernels of the edge detection algorithm are applied in strides to the pixels of the image ( 13 ).

8. A system ( 10 ) for assigning a label of one of a plurality of predetermined classes to each pixel of an image ( 13 ), the system ( 10 ) comprising:

a camera ( 11 ) capturing the image ( 13 ) and

a device ( 12 ) configured to:

receive the image ( 13 ) captured by the camera ( 11 ), the image ( 13 ) comprising a plurality of pixels;

use an encoder convolutional neural network to generate probability values for each pixel, each probability value indicating the probability that the respective pixel is associated with one of the plurality of predetermined classes;

generate for each pixel a class prediction value from the probability values, the class prediction value predicting the class of the plurality of predetermined classes the respective pixel is associated with;

use an edge detection algorithm to predict boundaries between objects shown in the image ( 13 ), the class prediction values of the pixels being used as input values of the edge detection algorithm; and

assign a label of one of the plurality of predetermined classes to each pixel of the image ( 13 ) based on the predicted boundaries.

9. A method ( 20 ) for assigning a label of one of a plurality of predetermined classes to each pixel of an image, the method comprising:

receiving an image ( 13 ) captured by a camera ( 11 ), the image ( 13 ) comprising a plurality of pixels;

using an encoder convolutional neural network to generate probability values for each pixel, each probability value indicating the probability that the respective pixel is associated with one of the plurality of predetermined classes;

generating for each pixel a class prediction value from the probability values, the class prediction value predicting the class of the plurality of predetermined classes the respective pixel is associated with;

using an edge detection algorithm to predict boundaries between objects shown in the image, the class prediction values of the pixels being used as input values of the edge detection algorithm; and

assigning a label of one of the plurality of predetermined classes to each pixel of the image ( 13 ) based on the predicted boundaries.

10. The method ( 20 ) as claimed in claim 9 , further comprising:

creating a mask ( 30 ) that covers pixels of the predicted boundaries; and

filtering the pixels of the predicted boundaries covered by the mask ( 30 ).

11. The method ( 20 ) as claimed in claim 10 , further comprising adding pixels in local neighborhoods of the predicted boundaries to the mask ( 30 ).

12. The method ( 20 ) as claimed in claim 9 , wherein for each pixel the class predicting value is generated by selecting the class associated with the probability value having a highest probability value for the respective pixel.

13. The method ( 20 ) as claimed in claim 9 , wherein the camera ( 11 ) is mounted on a vehicle.

14. The method ( 20 ) as claimed in claim 9 , wherein the edge detection algorithm is a Sobel edge detection algorithm.

15. The method ( 20 ) as claimed in claim 9 , wherein kernels of the edge detection algorithm are applied in strides to the pixels of the image ( 13 ).