Granted Patent B2

US 12,505,521 · App. 17/806,804 · Granted Dec 23, 2025

Inventors: Derek J. Phillips (Mountain View, CA); David Peng (Wilmette, IL)

Assignee: Kodiak Robotics, Inc.

G06T5/80G06T7/75G06T2207/10028G06T2207/20024

Abstract

Claims (73)

1 . A method of filtering atmospheric conditions from Light Detection and Ranging (LiDAR) point clouds, comprising:

generating, using a LiDAR system, at least one point cloud, wherein the LiDAR system includes a processor; and

using the processor:

identifying and isolating one or more ground points within a point cloud of the at least one point cloud, wherein the one or more ground points indicate a ground portion within an environment of the point cloud;

filtering out the ground portion from the point cloud, generating an initial processed point cloud;

identifying and isolating one or more atmospheric condition points within the initial processed point cloud, wherein the one or more atmospheric condition points indicate one or more atmospheric conditions within an environment of the initial processed point cloud,

wherein the identifying and isolating the one or more atmospheric condition points comprises:

identifying one or more high-intensity points and one or more low-intensity points, and a position for each of the one or more high-intensity point and each of the one or more low-intensity points, within the initial processed point cloud;

for each of the one or more low-intensity points, determining a mean distance of the low-intensity point to one or more of the one or more high-intensity points; and

for each of the one or more low-intensity points, when the mean distance is greater than a threshold distance, classifying the low-intensity point as an atmospheric condition point; and

filtering out the atmospheric condition points from the initial processed point cloud, generating a final processed point cloud.

2 . The method of claim 1 , wherein the identifying the one or more high-intensity points and the one or more low-intensity points within the initial processed point cloud comprises:

determining an intensity score for each point in the initial processed point cloud;

classifying each point in the initial processed point cloud having an intensity score above an intensity threshold as a high-intensity point; and

classifying each point in the initial processed point cloud having an intensity score below the intensity threshold a low-intensity point.

3 . The method of claim 1 , further comprising classifying all points not filtered out in the final processed point cloud as obstacle points, wherein the obstacle points indicate one or more objects within the environment of the final processed point cloud.

4 . The method of claim 1 , wherein the one or more atmospheric conditions comprise rain or fog.

5 . The method of claim 1 , wherein:

the at least one point cloud includes an initial point cloud and a subsequent point cloud; and

the identifying and isolating the one or more ground points comprises:

identifying and isolating one or more ground points within the initial point cloud; and

identifying and isolating one or more ground points within the subsequent point cloud by comparing the one or more ground points within the initial point cloud against one or more points within the subsequent point cloud.

6 . The method of claim 1 , wherein the identifying and isolating the one or more ground points comprises comparing the point cloud against one or more secondary scans of the environment of the point cloud.

7 . The method of claim 6 , wherein the one or more secondary scans include one or more of: one or more 2-dimensional camera images of the environment; and one or more radio detection and ranging (RADAR) scans of the environment.

8 . A system for filtering atmospheric conditions from Light Detection and Ranging (LiDAR) point clouds, comprising:

a vehicle; and

a LiDAR system, coupled to the vehicle, the LiDAR system comprising:

one or more LiDAR sensors; and

a processor, configured to:

generate at least one point cloud of an environment;

identify and isolate one or more ground points within a point cloud of the at least one point cloud, wherein the one or more ground points indicate a ground portion within an environment of the point cloud;

filter out the ground portion from the point cloud, generating an initial processed point cloud;

identify and isolate one or more atmospheric condition points within the initial processed point cloud, wherein the one or more atmospheric condition points indicate one or more atmospheric conditions within an environment of the initial processed point cloud,

wherein the identifying and isolating the one or more atmospheric condition points comprises:

identifying one or more high-intensity points and one or more low-intensity points, and a position for each of the one or more high-intensity point and each of the one or more low-intensity points, within the initial processed point cloud;

for each of the one or more low-intensity points, determining a mean distance of the low-intensity point to one or more of the one or more high-intensity points; and

for each of the one or more low-intensity points, when the mean distance is greater than a threshold distance, classifying the low-intensity point as an atmospheric condition point; and

filter out the atmospheric condition points from the initial processed point cloud, generating a final processed point cloud.

9 . The system of claim 8 , wherein the identifying the one or more high-intensity points and the one or more low-intensity points within the initial processed point cloud comprises:

determining an intensity score for each point in the initial processed point cloud;

classifying each point in the initial processed point cloud having an intensity score above an intensity threshold as a high-intensity point; and

classifying each point in the initial processed point cloud having an intensity score below the intensity threshold a low-intensity point.

10 . The system of claim 8 , wherein the processor is further configured to classify all points not filtered out in the final processed point cloud as obstacle points, wherein the obstacle points indicate one or more objects within the environment of the final processed point cloud.

11 . The system of claim 8 , wherein:

the at least one point cloud includes an initial point cloud and a subsequent point cloud; and

the identifying and isolating the one or more ground points comprises:

identifying and isolating one or more ground points within the initial point cloud; and

identifying and isolating one or more ground points within the subsequent point cloud by comparing the one or more ground points within the initial point cloud against one or more points within the subsequent point cloud.

12 . The system of claim 8 , further comprising one or more secondary scanners configured to:

scan the environment of the point cloud; and

generate one or more secondary scans of the environment of the point cloud,

wherein the identifying and isolating the one or more ground points comprises comparing the point cloud against the one or more secondary scans of the environment of the point cloud.

13 . The system of claim 12 , wherein the one or more secondary scanners include one or more of: one or more cameras; and one or more radio detection and ranging (RADAR) scanners.

14 . A system, comprising:

at least one Light Detection and Ranging (LiDAR) system, coupled to a vehicle, configured to generate one or more point clouds of an environment; and

a computing device, including a processor and a memory, coupled to the vehicle, configured to store programming instructions that, when executed by the processor, cause the processor to:

identify and isolate one or more ground points within a point cloud of the one or more point clouds, wherein the one or more ground points indicate a ground portion within an environment of the point cloud;

filter out the ground portion from the point cloud, generating an initial processed point cloud;

identify and isolate one or more atmospheric condition points within the initial processed point cloud, wherein the one or more atmospheric condition points indicate one or more atmospheric conditions within an environment of the initial processed point cloud,

wherein the identifying and isolating the one or more atmospheric condition points comprises:

identifying one or more high-intensity points and one or more low-intensity points, and a position for each of the one or more high-intensity point and each of the one or more low-intensity points, within the initial processed point cloud;

for each of the one or more low-intensity points, determining a mean distance of the low-intensity point to one or more of the one or more high-intensity points; and

for each of the one or more low-intensity points, when the mean distance is greater than a threshold distance, classifying the low-intensity point as an atmospheric condition point; and

filter out the atmospheric condition points from the initial processed point cloud, generating a final processed point cloud.

15 . The system of claim 14 , wherein, when the processor identifies the one or more high-intensity points and the one or more low-intensity points within the initial processed point cloud, the programming instructions are further configured to cause the processor to:

determine an intensity score for each point in the initial processed point cloud;

classify each point in the initial processed point cloud having an intensity score above an intensity threshold as a high-intensity point; and

classify each point in the initial processed point cloud having an intensity score below the intensity threshold a low-intensity point.

16 . The system of claim 14 , wherein the programming instructions are further configured to cause the processor to classify all points not filtered out in the final processed point cloud as obstacle points, wherein the obstacle points indicate one or more objects within the environment of the final processed point cloud.

17 . The system of claim 14 , wherein:

the one or more point clouds include an initial point cloud and a subsequent point cloud; and

when the processor identifies and isolates the one or more ground points, the programming instructions are further configured to cause the processor to:

identify and isolate one or more ground points within the initial point cloud; and identify and isolate one or more ground points within the subsequent point cloud by comparing the one or more ground points within the initial point cloud against one or more points within the subsequent point cloud.

Assignments (4)

Continuity (1)

References Cited (15)