Parametric top-view representation of scenes

1. A computer-implemented method for implementing parametric models for scene representation to improve autonomous task performance, comprising:

generating an initial map of a scene based on at least one image corresponding to a perspective view of the scene, the initial map including a non-parametric top-view representation of the scene, the generating the initial map further includes performing occlusion-reasoning by:

extracting a feature representation including one or more features from the at least one image;

using semantic segmentation to obtain a semantic category for each of a plurality of pixels of the at least one image based on the feature representation;

generating a masked feature representation based on the semantic segmentation, including masking out one or more regions corresponding to one or more foreground objects identified in the scene;

predicting a background category and a depth value for each of the plurality of pixels based on the masked feature representation; and

mapping each of the plurality of pixels to three-dimensions based on the background categories and the depth values to generate the initial map;

implementing a parametric model to obtain a scene element representation based on the initial map, the scene element representation providing a description of one or more scene elements of the scene and corresponding to an estimated semantic layout of the scene;

identifying one or more predicted locations of the one or more scene elements by performing three-dimensional localization based on the at least one image; and

obtaining an overlay for performing an autonomous task by placing the one or more scene elements with the one or more respective predicted locations onto the scene element representation.

2. The method of claim 1 , further comprising obtaining the at least one image using an image capturing device, wherein mapping each of the plurality of pixels to three-dimensions further includes using an intrinsic calibration of the image capturing device.

3. The method of 2 , wherein the image capturing device is mounted on a vehicle such that the at least one image includes a perspective view of a driving scene.

4. The method of claim 1 , wherein implementing the parametric model to obtain the scene element representation further includes implementing a graphical model to ensure feasibility of the scene element representation based on a probability corresponding to each of one or more scene parameters obtained from the initial map.

5. The method of claim 1 , further comprising performing the autonomous task by applying the overlay to one or more applications, wherein the one or more applications include one or more of free-space estimation to compute free space, path planning for computation of distances, and behavior prediction.

6. The method of claim 1 , further comprising training the parametric model based on the initial map.

7. A 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 for implementing parametric models for scene representation to improve autonomous task performance, the method performed by the computer comprising:

generating an initial map of a scene based on at least one image corresponding to a perspective view of the scene, the initial map including a non-parametric top-view representation of the scene, the generating the initial map further includes performing occlusion-reasoning by:

extracting a feature representation including one or more features from the at least one image;

using semantic segmentation to obtain a semantic category for each of a plurality of pixels of the at least one image based on the feature representation;

generating a masked feature representation based on the semantic segmentation, including masking out one or more regions corresponding to one or more foreground objects identified in the scene;

predicting a background category and a depth value for each of the plurality of pixels based on the masked feature representation; and

mapping each of the plurality of pixels to three-dimensions based on the background categories and the depth values to generate the initial map;

implementing a parametric model to obtain a scene element representation based on the initial map, the scene element representation providing a description of one or more scene elements of the scene and corresponding to an estimated semantic layout of the scene;

identifying one or more predicted locations of the one or more scene elements by performing three-dimensional localization based on the at least one image; and

obtaining an overlay for performing an autonomous task by placing the one or more scene elements with the one or more respective predicted locations onto the scene element representation.

8. The computer program product of claim 7 , wherein the method further includes obtaining the at least one image using an image capturing device, and wherein mapping each of the plurality of pixels to three-dimensions further includes using an intrinsic calibration of the image capturing device.

9. The computer program product of claim 8 , wherein the image capturing device is mounted on a vehicle such that the at least one image includes a perspective view of a driving scene.

10. The computer program product of claim 7 , wherein implementing the parametric model to obtain the scene element representation further includes implementing a graphical model to ensure feasibility of the scene element representation based on a probability corresponding to each of one or more scene parameters.

11. The computer program product of claim 7 , wherein the method further includes performing the autonomous task by applying the overlay to one or more applications, wherein the one or more applications include one or more of free-space estimation to compute free space, path planning for computation of distances, and behavior prediction.

12. The computer program product of claim 7 , wherein the method further includes training the parametric model based on the initial map.

13. A system for implementing parametric models for scene representation to improve autonomous task performance, comprising:

a memory device for storing program code; and

at least one processor device operatively coupled to a memory device and configured to execute program code stored on the memory device to:

generate an initial map of a scene based on at least one image corresponding to a perspective view of the scene, the initial map including a non-parametric top-view representation of the scene, the initial map being generated by performing occlusion-reasoning by:

extracting a feature representation including one or more features from the at least one image;

using semantic segmentation to obtain a semantic category for each of a plurality of pixels of the at least one image based on the feature representation;

generating a masked feature representation based on the semantic segmentation, including masking out one or more regions corresponding to one or more foreground objects identified in the scene;

predicting a background category and a depth value for each of the plurality of pixels based on the masked feature representation; and

mapping each of the plurality of pixels to three-dimensions based on the background categories and the depth values to generate the initial map;

implement a parametric model to obtain a scene element representation based on the initial map, the scene element representation providing a description of one or more scene elements of the scene and corresponding to an estimated semantic layout of the scene;

identify one or more predicted locations of the one or more scene elements by performing three-dimensional localization based on the at least one image; and

obtain an overlay for performing an autonomous task by placing the one or more scene elements with the one or more respective predicted locations onto the scene element representation.

14. The system of claim 13 , further comprising an image capturing device, wherein the at least one processor device is further configured to execute program code stored on the memory device to obtain the at least one image using the image capturing device, and wherein the at least one processor device is further configured to map each of the plurality of pixels to three-dimensions by using an intrinsic calibration of the image capturing device.

15. The system of claim 14 , wherein the image capturing device is mounted on a vehicle such that the at least one image includes a perspective view of a driving scene.

16. The system of claim 13 , wherein the at least once processor device is further configured to implement the parametric model to obtain the scene element representation by implementing a graphical model to ensure feasibility of the scene element representation based on a probability corresponding to each of one or more scene parameters.

17. The system of claim 13 , wherein the at least one processor device is further configured to execute program code stored on the memory device to perform the autonomous task by applying the overlay to one or more applications, wherein the one or more applications include one or more of free-space estimation to compute free space, path planning for computation of distances, and behavior prediction.