Rut detection for road infrastructure
A computer-implemented method for rut detection is provided. The method includes detecting, by a rut detection system, areas in a road-scene image that include ruts with pixel-wise probability values, wherein a higher value indicates a better chance of being a rut. The method further includes performing at least one of rut repair and vehicle rut avoidance responsive to the pixel-wise probability values. The detecting step includes performing neural network-based, pixel-wise semantic segmentation with context information on the road-scene image to distinguish rut pixels from non-rut pixels on a road depicted in the road-scene image.
1. A computer-implemented method for rut detection, comprising:
detecting, by a rut detection system, areas in a road-scene image that include ruts with pixel-wise probability values, wherein a higher value indicates a better chance of being a rut; and
performing at least one of rut repair and vehicle rut avoidance responsive to the pixel-wise probability values,
wherein said detecting step comprises performing neural network-based, pixel-wise semantic segmentation with context information on the road-scene image to distinguish rut pixels from non-rut pixels on a road depicted in the road-scene image.
2. The computer-implemented method of claim 1 , wherein the context information includes context categories comprising road, car, ruts, and cracks.
3. The computer-implemented method of claim 1 , further comprising training a neural network performing the neural network-based, pixel-wise semantic segmentation with (a) pixel-wise ground truths for labeled data pertaining to non-rut categories and (b) pixel-wise annotations for rut categories comprising ruts and cracks.
4. The computer-implemented method of claim 3 , further comprising limiting annotation costs during the training using unlabeled data or data with weak annotations, the weak annotations being image-wise labels compared to image-wise ground truths for the unlabeled data or the data with the weak annotations.
5. The computer-implemented method of claim 3 , wherein the training comprises using the pixel-wise ground truths as a supervision to optimize a task loss for pixel-wise rut detection.
6. The computer-implemented method of claim 1 , wherein the neural network-based, pixel-wise semantic segmentation comprises a segmentation convolutional neural network.
7. The computer-implemented method of claim 1 , wherein said detecting step uses a weak loss for image classification with respect to an image-wise classification generated for unlabeled data or data with weak annotations comprising image-wise annotations.
8. The computer-implemented method of claim 1 , wherein the vehicle rut avoidance involves controlling a vehicle system directed to at least one of steering, accelerating, and braking, responsive to the pixel-wise probability values.
9. A computer program product for rut detection, the computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method comprising:
detecting, by a rut detection system, areas in a road-scene image that include ruts with pixel-wise probability values, wherein a higher value indicates a better chance of being a rut; and
performing at least one of rut repair and vehicle rut avoidance responsive to the pixel-wise probability values,
wherein said detecting step comprises performing neural network-based, pixel-wise semantic segmentation with context information on the road-scene image to distinguish rut pixels from non-rut pixels on a road depicted in the road-scene image.
10. The computer program product of claim 9 , wherein the context information includes context categories comprising road, car, ruts, and cracks.
11. The computer program product of claim 9 , further comprising training a neural network performing the neural network-based, pixel-wise semantic segmentation with (a) pixel-wise ground truths for labeled data pertaining to non-rut categories and (b) pixel-wise annotations for rut categories comprising ruts and cracks.
12. The computer program product of claim 11 , further comprising limiting annotation costs during the training using unlabeled data or data with weak annotations, the weak annotations being image-wise labels compared to image-wise ground truths for the unlabeled data or the data with the weak annotations.
13. The computer program product of claim 11 , wherein the training comprises using the pixel-wise ground truths as a supervision to optimize a task loss for pixel-wise rut detection.
14. The computer program product of claim 9 , wherein the neural network-based, pixel-wise semantic segmentation comprises a segmentation convolutional neural network.
15. The computer program product of claim 9 , wherein said detecting step uses a weak loss for image classification with respect to an image-wise classification generated for unlabeled data or data with weak annotations comprising image-wise annotations.
16. The computer program product of claim 9 , wherein the vehicle rut avoidance involves controlling a vehicle system directed to at least one of steering, accelerating, and braking, responsive to the pixel-wise probability values.
17. A rut detection and repair system, comprising:
a camera for capturing a road-scene image;
a memory device for storing program code therein;
a processor device operatively coupled to the memory device for running the program code to
detect areas in the road-scene image that include ruts with pixel-wise probability values, wherein a higher value indicates a better chance of being a rut; and
control one or more physical machines performing rut repair responsive to the pixel-wise probability values,
wherein the processor device detects areas in the road-scene image by performing neural network-based, pixel-wise semantic segmentation with context information on the road-scene image to distinguish rut pixels from non-rut pixels on a road depicted in the road-scene image.
18. The rut detection and repair system of claim 17 , wherein the context information includes context categories comprising road, car, ruts, and cracks.
19. The rut detection and repair system of claim 17 , wherein the processor device further runs the program code to train a neural network performing the neural network-based, pixel-wise semantic segmentation with (a) pixel-wise ground truths for labeled data pertaining to non-rut categories and (b) pixel-wise annotations for rut categories comprising ruts and cracks.
20. The rut detection and repair system of claim 19 , wherein the processor device further runs the program code to limit annotation costs during the training using unlabeled data or data with weak annotations, the weak annotations being image-wise labels compared to image-wise ground truths for the unlabeled data or the data with the weak annotations.