Robust simultaneous localization and mapping via removal of dynamic traffic participants

1. A method for localization and mapping for an autonomous vehicle, comprising:

obtaining, via a camera, an image acquired at a first location;

assigning labels to one or more objects of the image;

obtaining, via a light detector, a point cloud acquired at a second location, which is within a threshold distance of the first location;

determining one or more object clusters of the point cloud;

associating the labels assigned to the one or more objects of the image with points of the object clusters of the point cloud; and

identifying three-dimensional (3D) objects of the point cloud based on the labels associated with the points of the object clusters.

2. The method of claim 1 , further comprising:

removing dynamic traffic participants from the point cloud based on the identified 3D objects.

3. The method of claim 2 , further comprising:

performing a simultaneous localization and mapping operation on the point cloud after the dynamic traffic participants are removed from the point cloud.

4. The method of claim 1 , wherein the assigning labels to the one or more objects of the image includes performing semantic segmentation to the image on a per pixel basis.

5. The method of claim 4 , wherein the associating the labels includes associating labels of at least a portion of pixels of the image to corresponding points of the point cloud.

6. The method of claim 1 , wherein the assigning labels to one or more objects of the image includes identifying the dynamic traffic participants within the image.

7. The method of claim 1 , wherein the determining one or more object clusters of the point cloud includes identifying a ground plane of the point cloud and removing the ground plane from the point cloud.

8. A system for localization and mapping for an autonomous vehicle, comprising:

a camera for capturing one or more images;

a light detector for capturing one or more point clouds;

a memory coupled to the camera and the light detector for storing the one or more images and the one or more point clouds; and

one or more processors coupled to the memory, wherein the one or more processors is configured to:

obtain, via the camera, an image acquired at a first location;

assign labels to one or more objects of the image;

obtain, via the light detector, a point cloud acquired at a second location, which is within a threshold distance of the first location;

determine one or more object clusters of the point cloud;

associate the labels assigned to the one or more objects of the image with points of the object clusters of the point cloud; and

identify three-dimensional (3D) objects of the point cloud based on the labels associated with the points of the object clusters.

9. The system of claim 8 , wherein the one or more processors is further configured to:

remove dynamic traffic participants from the point cloud based on the identified 3D objects.

10. The system of claim 9 , wherein the one or more processors is further configured to:

perform a simultaneous localization and mapping operation on the point cloud after the dynamic traffic participants are removed from the point cloud.

11. The system of claim 8 , wherein the one or more processors is further configured to:

perform semantic segmentation to the image on a per pixel basis.

12. The system of claim 11 , wherein the one or more processors is further configured to:

associate labels of at least a portion of pixels of the image to corresponding points of the point cloud.

13. The system of claim 8 , wherein the one or more processors is further configured to:

identify the dynamic traffic participants within the image.

14. The system of claim 8 , wherein the one or more processors is further configured to:

identify a ground plane of the point cloud; and

remove the ground plane from the point cloud.

15. A non-transitory computer-readable medium storing computer executable code for one or more processors to perform localization and mapping for an autonomous vehicle, comprising code for:

obtaining, via a camera, an image acquired at a first location;

assigning labels to one or more objects of the image;

obtaining, via a light detector, a point cloud acquired at a second location, which is within a threshold distance of the first location;

determining one or more object clusters of the point cloud;

associating the labels assigned to the one or more objects of the image with points of the object clusters of the point cloud; and

identifying three-dimensional (3D) objects of the point cloud based on the labels associated with the points of the object clusters.

16. The non-transitory computer-readable medium of claim 15 , further comprising code for:

removing dynamic traffic participants from the point cloud based on the identified 3D objects.

17. The non-transitory computer-readable medium of claim 16 , further comprising code for:

performing a simultaneous localization and mapping operation on the point cloud after the dynamic traffic participants are removed from the point cloud.

18. The non-transitory computer-readable medium of claim 15 , wherein the code for assigning labels to the one or more objects of the image includes code for performing semantic segmentation to the image on a per pixel basis.

19. The non-transitory computer-readable medium of claim 18 , wherein the code for associating the labels includes code for associating labels of at least a portion of pixels of the image to corresponding points of the point cloud.

20. The non-transitory computer-readable medium of claim 15 , wherein the code for assigning labels to one or more objects of the image includes code identifying the dynamic traffic participants within the image.