System and method for self-supervised monocular ground-plane extraction
A method for controlling an agent to navigate through an environment includes generating a depth map associated with a monocular image of the environment. The method also includes generating a group of surface normal. Each surface normal of the group of surface normals is associated with a respective polygon of a group of polygons associated with the depth map. The method further includes identifying one or more ground planes in the depth map based on the group of surface normal. The method further includes controlling the agent to navigate through the environment based on identifying the one or more ground planes.
1. An apparatus for controlling an agent to navigate through an environment, comprising:
one or more processors; and
one or more memories coupled with the one or more processors and storing processor-executable code that, when executed by the one or more processors, is configured to cause the apparatus to:
generate a first depth map and a second depth map associated with a monocular image of the environment, the first depth map having a first resolution and the second depth map having a second resolution that is different than the first resolution;
select one of the first depth map or the second depth map for a ground plane extraction task;
generate a plurality of surface normals, each surface normal of the plurality of surface normals being a vector cross product of two edges from a respective polygon of a plurality of polygons associated with the selected depth map;
identify a group of planes in the selected depth map based on the plurality of surface normals, each one of the group of planes being associated with a respective set of surface normals, from the plurality of surface normals, facing a same direction, at least one plane of the group of planes indicating a ground plane, at least one plane of the group of planes being a second plane that is different from the ground plane;
generate a height map of the environment based on the identified group of planes; and
control the agent to navigate through the environment based on generating the height map.
2. The apparatus of claim 1 , wherein each of the plurality of polygons is a triangle.
3. The apparatus of claim 1 , wherein the set of surface normal associated with the ground plane are perpendicular with respect to a direction of the agent and/or a sensor associated with the agent.
4. The apparatus of claim 1 , wherein the ground plane corresponds to a road in the environment.
5. The apparatus of claim 1 , wherein:
the agent is an autonomous or semi-autonomous vehicle; and
the monocular image is captured via a sensor integrated with the agent.
6. The apparatus of claim 1 , wherein execution of the processor-executable code further causes the apparatus to train a depth model, in a self-supervising manner, on the first depth map, the second depth map, and the one or more ground planes.
7. A method for controlling an agent to navigate through an environment, comprising:
generating a first depth map and a second depth map associated with a monocular image of the environment, the first depth map having a first resolution and the second depth map having a second resolution that is different than the first resolution;
selecting one of the first depth map or the second depth map for a ground plane extraction task;
generating a plurality of surface normals, each surface normal of the plurality of surface normals being a vector cross product of two edges from a respective polygon of a plurality of polygons associated with the selected depth map;
identifying a group of planes in the selected depth map based on the plurality of surface normals, each one of the group of planes being associated with a respective set of surface normals, from the plurality of surface normals, facing a same direction, at least one plane of the group of planes indicating a ground plane, at least one plane of the group of planes being a second plane that is different from the ground plane;
generating a height map of the environment based on the identified group of planes; and
controlling the agent to navigate through the environment based on generating the height map.
8. The method of claim 7 , wherein each of the plurality of polygons is a triangle.
9. The method of claim 7 , wherein the set of surface normal associated with the ground plane are perpendicular with respect to a direction of the agent and/or a sensor associated with the agent.
10. The method of claim 7 , wherein the ground plane corresponds to a road in the environment.
11. The method of claim 7 , wherein:
the agent is an autonomous or semi-autonomous vehicle; and
the monocular image is captured via a sensor integrated with the agent.
12. The method of claim 7 , further comprising training a depth model, in a self-supervising manner, on the first depth map, the second depth map, and the one or more ground planes.
13. A non-transitory computer-readable medium having program code recorded thereon for controlling an agent to navigate through an environment, the program code executed by a processor and comprising:
program code to generate a first depth map and a second depth map associated with a monocular image of the environment, the first depth map having a first resolution and the second depth map having a second resolution that is different than the first resolution;
program code to select one of the first depth map or the second depth map for a ground plane extraction task;
program code to generate a plurality of surface normals, each surface normal of the plurality of surface normals being a vector cross product of two edges from a respective polygon of a plurality of polygons associated with the selected depth map;
program code to identify a group of planes in the selected depth map based on the plurality of surface normals, each one of the group of planes being associated with a respective set of surface normals, from the plurality of surface normals, facing a same direction, at least one plane of the group of planes indicating a ground plane, at least one plane of the group of planes being a second plane that is different from the ground plane;
program code to generate a height map of the environment based on the identified group of planes; and
program code to control the agent to navigate through the environment based on generating the height map.
14. The non-transitory computer-readable medium of claim 13 , wherein each of the plurality of polygons is a triangle.
15. The non-transitory computer-readable medium of claim 13 , wherein the ground plane corresponds to a road in the environment.
16. The non-transitory computer-readable medium of claim 13 , wherein:
the agent is an autonomous or semi-autonomous vehicle; and
the monocular image is captured via a sensor integrated with the agent.
17. The non-transitory computer-readable medium of claim 13 , wherein the set of surface normal associated with the ground plane are perpendicular with respect to a direction of the agent and/or a sensor associated with the agent.
18. The non-transitory computer-readable medium of claim 13 , wherein the program code further comprises program code to train a depth model, in a self-supervising manner, on the first depth map, the second depth map, and the one or more ground planes.