End-to-end parametric road layout prediction with cheap supervision
A computer-implemented method for road layout prediction is provided. The method includes segmenting, by a first processor-based element, an RGB image to output pixel-level semantic segmentation results for the RGB image in a perspective view for both visible and occluded pixels in the perspective view based on contextual clues. The method further includes learning, by a second processor-based element, a mapping from the pixel-level semantic segmentation results for the RGB image in the perspective view to a top view of the RGB image using a road plane assumption. The method also includes generating, by a third processor-based element, an occlusion-aware parametric road layout prediction for road layout related attributes in the top view.
1. A computer-implemented method for road layout prediction, comprising:
generating, by performing occlusion-reasoned pixel-level semantic segmentation to an image in a perspective view of a road, segmented results including semantics of both visible and occluded pixels in the perspective view;
mapping from the segmented results including semantics of both visible and occluded pixels in the perspective view to a top view using a road plane assumption; and
generating, based on the segmented results mapped to the top view including semantics of both visible and occluded pixels, a parametric prediction for road layout related attributes in the top view.
2. The computer-implemented method of claim 1 , wherein the step of the generating of the parametric prediction comprises receiving pixel-level human annotations only for the road layout related attributes in the top view, and wherein the step of the generating of the segmented results and the step of the learning of the mapping are performed without human annotations.
3. The computer-implemented method of claim 1 , wherein the step of the generating of the segmented results and the step of the learning of the mapping use pixel-level supervision signals.
4. The computer-implemented method of claim 1 , wherein the parametric prediction for the road layout related attributes comprises a prediction boundary selected from a group consisting of a road boundary, a crosswalk boundary, a sidewalk boundary, and a lane boundary, wherein one or more members of the group are originally occluded by a respective foreground element in the perspective view.
5. The computer-implemented method of claim 1 , further comprising controlling a vehicle system of a vehicle for accident avoidance responsive to the parametric prediction for the road layout related attributes.
6. The computer-implemented method of claim 5 , wherein the vehicle system is selected from a group consisting of a braking system, an accelerating system, a stability system, and a steering system.
7. The computer-implemented method of claim 1 , wherein the method is performed by an Advanced Driver Assistance System.
8. The computer-implemented method of claim 1 , further comprising recovering noisy semantics in the segmented results mapped to the top view including semantics of both visible and occluded pixels.
9. A computer program product for road layout prediction, 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:
generating, by performing occlusion-reasoned pixel-level semantic segmentation to an image in a perspective view of a road, segmented results including semantics of both visible and occluded pixels in the perspective view;
mapping from the segmented results including semantics of both visible and occluded pixels in the perspective view to a top view of using a road plane assumption; and
generating, based on the segmented results mapped to the top view including semantics of both visible and occluded pixels, a parametric prediction for road layout related attributes in the top view.
10. The computer program product of claim 9 , wherein the step of the generating of the parametric prediction comprises receiving pixel-level human annotations only for the road layout related attributes in the top view, and wherein the step of the generating of the segmented results and the step of the learning of the mapping are performed without human annotations.
11. The computer program product of claim 9 , wherein the step of the generating of the segmented results and the step of the learning of the mapping use pixel-level supervision signals.
12. The computer program product of claim 9 , wherein the parametric prediction for the road layout related attributes comprises a prediction boundary selected from a group consisting of a road boundary, a crosswalk boundary, a sidewalk boundary, and a lane boundary, wherein one or more members of the group are originally occluded by a respective foreground element in the perspective view.
13. The computer program product of claim 9 , wherein the method further comprises controlling a vehicle system of a vehicle for accident avoidance responsive to the parametric prediction for the road layout related attributes.
14. The computer program product of claim 13 , wherein the vehicle system is selected from a group consisting of a braking system, an accelerating system, a stability system, and a steering system.
15. A computer processing system for road layout prediction, comprising:
a memory device for storing program code; and
a processor device for running the program code to:
generate, by performing occlusion-reasoned pixel-level semantic segmentation to an image in a perspective view of a road, segmented results including semantics of both visible and occluded pixels in the perspective view;
map from the segmented results including semantics of both visible and occluded pixels in the perspective view to a top view using a road plane assumption; and
generate, based on the segmented results mapped to the top view including semantics of both visible and occluded pixels, a parametric prediction for road layout related attributes in the top view.
16. The computer-implemented method of claim 8 , further comprising generating, from the segmented results in the top view in which the noisy semantics are recovered, an image of the perspective view in which occluded regions are recovered.