Segmenting ground points from non-ground points to assist with localization of autonomous vehicles
According to an aspect of an embodiment, operations may comprise receiving, from a LIDAR mounted on a vehicle, a first 3D point cloud comprising points of a region around the vehicle as observed by the LIDAR. The operations may also comprise accessing an HD map comprising a second 3D point cloud comprising points of the region around the vehicle. The operations may also comprise segmenting LIDAR ground points from LIDAR non-ground points in the first 3D point cloud. The operations may also comprise segmenting map ground points from map non-ground points in the second 3D point cloud. The operations may also comprise determining a pose of the vehicle by matching the LIDAR ground points to the map ground points and by matching the LIDAR non-ground points to the map non-ground points.
1. A method, comprising:
receiving a first three dimensional (3D) point cloud comprising points corresponding to a region around a vehicle as observed by a light detection and ranging (LIDAR) sensor mounted on the vehicle;
accessing a high definition (HD) map comprising a second 3D point cloud comprising points corresponding to the region around the vehicle;
segmenting LIDAR ground points from LIDAR non-ground points in the first 3D point cloud;
segmenting map ground points from map non-ground points in the second 3D point cloud;
determining a first correspondence between one or more LIDAR ground points and one or more map ground points;
determining a second correspondence between one or more LIDAR non-ground points and one or more map non-ground points; and
determining a pose of the vehicle based at least on the first correspondence and the second correspondence.
2. The method of claim 1 , wherein the segmenting of the LIDAR ground points from the LIDAR non-ground points in the first 3D point cloud and the segmenting of the map ground points from the map non-ground points in the second 3D point cloud are performed using a random sample consensus (RANSAC) algorithm.
3. The method of claim 1 , wherein the segmenting of the LIDAR ground points from the LIDAR non-ground points in the first 3D point cloud and the segmenting of the map ground points from the map non-ground points in the second 3D point cloud are performed using a range-based segmentation algorithm.
4. The method of claim 1 , wherein the determining the first correspondence is performed separately from the determining the second correspondence.
5. The method of claim 1 , wherein a different weight is applied to the first correspondence than to the second correspondence.
6. The method of claim 1 , wherein:
a roll, a pitch, and an altitude of the pose is determined by performing an iterative closet point (ICP) algorithm to match the one or more LIDAR ground points to the one or more map ground points; and
an x coordinate, a y coordinate, and a yaw of the pose is determined by performing the ICP algorithm to match the one or more LIDAR non-ground points to the one or more map non-ground points.
7. A processor comprising:
processing circuitry to cause performance of operations comprising:
receiving a first three dimensional (3D) point cloud comprising points corresponding to a region around a vehicle as observed by a light detection and ranging (LIDAR) sensor mounted on the vehicle;
accessing a high definition (HD) map comprising a second 3D point cloud comprising points corresponding to the region around the vehicle;
segmenting LIDAR ground points from LIDAR non-ground points in the first 3D point cloud;
segmenting map ground points from map non-ground points in the second 3D point cloud;
matching points of the first 3D point cloud to points of the second 3D point cloud, the matching including the LIDAR ground points only being compared against the map ground points and the LIDAR non-ground points only being compared against the map non-ground points; and
determining a pose of the vehicle based at least on the matching, wherein matches between the LIDAR ground points and the map ground points are weighted differently than matches between the LIDAR non-ground points and the map non-ground points in the determining the pose.
8. The processor of claim 7 , wherein the segmenting of the LIDAR ground points from the LIDAR non-ground points in the first 3D point cloud and the segmenting of the map ground points from the map non-ground points in the second 3D point cloud are performed using a random sample consensus (RANSAC) algorithm.
9. The processor of claim 7 , wherein the segmenting of the LIDAR ground points from the LIDAR non-ground points in the first 3D point cloud and the segmenting of the map ground points from the map non-ground points in the second 3D point cloud are performed using a range-based segmentation algorithm.
10. The processor of claim 7 , wherein the LIDAR ground points are compared to the map ground points separately from the LIDAR non-ground points being compared to the map non-ground points.
11. The processor of claim 7 , wherein a higher weight is applied to the matches between the LIDAR non-ground points and the map non-ground points than to the matches between the LIDAR ground points and the map ground points.
12. The processor of claim 7 , wherein:
a roll, a pitch, and an altitude of the pose is determined by performing an iterative closet point (ICP) algorithm to match the LIDAR ground points to the map ground points; and
an x coordinate, a y coordinate, and a yaw of the pose is determined by performing the ICP algorithm to match the LIDAR non-ground points to the map non-ground points.
13. A system comprising:
one or more processing units to perform operations comprising:
receiving a first three dimensional (3D) point cloud comprising points corresponding to a region around a vehicle as observed by a light detection and ranging (LIDAR) sensor mounted on the vehicle;
accessing a high definition (HD) map comprising a second 3D point cloud comprising points corresponding to the region around the vehicle;
segmenting LIDAR ground points from LIDAR non-ground points in the first 3D point cloud;
segmenting map ground points from map non-ground points in the second 3D point cloud;
determining a first correspondence between one or more LIDAR ground points and one or more map ground points;
determining a second correspondence between one or more LIDAR non-ground points and one or more map non-ground points; and
determining a pose of the vehicle based at least on the first correspondence and the second correspondence.
14. The system of claim 13 , wherein the segmenting of the LIDAR ground points from the LIDAR non-ground points in the first 3D point cloud and the segmenting of the map ground points from the map non-ground points in the second 3D point cloud are performed using a random sample consensus (RANSAC) algorithm.
15. The system of claim 13 , wherein the segmenting of the LIDAR ground points from the LIDAR non-ground points in the first 3D point cloud and the segmenting of the map ground points from the map non-ground points in the second 3D point cloud are performed using a range-based segmentation algorithm.
16. The system of claim 13 wherein the determining the first correspondence is performed separately from the determining the second correspondence.
17. The system of claim 13 , wherein a different weight is applied to the first correspondence than to the second correspondence.
18. The system of claim 13 , wherein:
a roll, a pitch, and an altitude of the pose is determined by performing an iterative closet point (ICP) algorithm to match the one or more LIDAR ground points to the one or more map ground points; and
an x coordinate, a y coordinate, and a yaw of the pose is determined by performing the ICP algorithm to match the one or more LIDAR non-ground points to the one or more map non-ground points.