Automated object annotation using fused camera/LiDAR data points
The present disclosure is directed to a computer system and techniques for automatically annotating objects in map data used for navigating an autonomous vehicle. Generally, the computer system is configured to obtain LiDAR data points for an environment around an autonomous vehicle, project the LiDAR data points onto image data, detect a target object in the image data, extract a subset of the LiDAR data points that corresponds to the detected target object, register the detected target object in map data if the extracted subset of the LiDAR data points satisfies registration criteria, and navigate the autonomous vehicle in the environment according to the map data.
1. A system comprising:
one or more computer processors; and
one or more storage media storing instructions which, when executed by the one or more computer processors, cause performance of operations comprising:
obtaining map data used for navigating a vehicle, wherein the map data identifies a plurality of objects in an environment of the vehicle;
obtaining image data of an image of the environment of the vehicle, wherein the image comprises one or more objects;
obtaining, using a LiDAR, LiDAR data points corresponding to the one or more objects;
projecting, using a processing circuit, the LiDAR data points onto the image data to generate fused data, the fused data comprising a LiDAR data portion and an image data portion;
detecting, using the processing circuit, a first target object and a second target object in the fused data;
extracting, using the processing circuit, a first subset of the LiDAR data points that corresponds to the detected first target object;
extracting, using the processing circuit, a second subset of the LiDAR data points that corresponds to the detected second target object;
automatically annotating, using the processing circuit, the map data, wherein the automatically annotating comprises:
in accordance with a determination that the extracted first subset of the LiDAR data points that corresponds to the detected first target object satisfies registration criteria associated with the detected first target object, automatically annotating the map data corresponding to the detected first target object, and
in accordance with a determination that the extracted second subset of the LiDAR data points that corresponds to the detected second target object does not satisfy the registration criteria associated with the detected second target object, not annotating map data corresponding to the detected second target object; and
navigating, using a control circuit, the vehicle in the environment according to the annotated map data.
2. The system of claim 1 , wherein obtaining the LiDAR data points includes obtaining the LiDAR data points while navigating, using the control circuit, the vehicle in the environment.
3. The system of claim 2 , wherein obtaining the LiDAR data points further includes:
obtaining pose data for the vehicle while navigating the vehicle in the environment; and
associating, using the processing circuit, the LiDAR data points with the pose data for the vehicle.
4. The system of claim 3 , wherein the pose data for the vehicle indicates a three-dimensional pose of the vehicle with respect to a geographical location in the environment and wherein the operations further comprise determining, using the processing circuit, a geographical location of the detected target object based on the three-dimensional pose of the vehicle.
5. The system of claim 1 , wherein projecting the LiDAR data points onto the image data includes fusing the LiDAR data points with image data that is obtained for the environment around the vehicle.
6. The system of claim 1 , wherein automatically annotating the detected first target object present: in the map data includes indicating a location of the detected first target object in the map data, and
the location of the detected first target object in the map data is based on a geographical location associated with the extracted subset of the LiDAR data points.
7. The system of claim 1 , further comprising determining, using the processing circuit, a three-dimensional pose of the detected first target object, wherein annotating the detected first target object present in the map data includes representing the detected target object in the map data having the three-dimensional pose.
8. The system of claim 1 , the operations further comprising:
determining, using the processing circuit, a three-dimensional pose of the detected first target object, wherein the registration criteria includes a criterion that is satisfied when the extracted subset of the LiDAR data points that corresponds to the detected first target object is associated with a distance that is less than a distance threshold.
9. The system of claim 1 , wherein the detected first target object is a traffic light and wherein detecting the first target object in the image data includes using machine learning to detect the first target object in the image data.
10. The system of claim 1 , wherein the annotating comprises adding a bounding box to the map data such that the bounding box is superimposed on the image data at a location corresponding to the detected first target object in the map data to provide a visual indication of the detected first target object.
11. A method, comprising:
obtaining map data used for navigating a vehicle, wherein the map data identifies a plurality of objects in an environment of the vehicle;
obtaining image data of an image of the environment of the vehicle, wherein the image comprises one or more objects;
obtaining, using a LiDAR, LiDAR data points corresponding to the one or more objects;
projecting, using a processing circuit, the LiDAR data points onto the image data to generate fused data, the fused data comprising a LiDAR data portion and an image data portion;
detecting, using the processing circuit, a first target object and a second target object in the fused data;
extracting, using the processing circuit, a first subset of the LiDAR data points that corresponds to the detected first target object;
extracting, using the processing circuit, a second subset of the LiDAR data points that corresponds to the detected second target object;
automatically annotating, using the processing circuit, the map data, wherein the automatically annotating comprises:
in accordance with a determination that the extracted first subset of the LiDAR data points that corresponds to the detected first target object satisfies registration criteria associated with the detected first target object, automatically annotating the map data corresponding to the detected first target object, and
in accordance with a determination that the extracted second subset of the LiDAR data points that corresponds to the detected second target object does not satisfy the registration criteria associated with detected the second target object, not annotating the map data corresponding to the detected second target object; and
navigating, using a control circuit, the vehicle in the environment according to the annotated map data.
12. The method of claim 11 , wherein obtaining the LiDAR data points includes obtaining the LiDAR data points while navigating, using the control circuit, the vehicle in the environment.
13. The method of claim 12 , wherein obtaining the LiDAR data points further includes:
obtaining pose data for the vehicle while navigating the vehicle in the environment; and
associating, using the processing circuit, the LiDAR data points with the pose data for the vehicle.
14. The method of claim 13 , wherein the pose data for the vehicle indicates a three-dimensional pose of the vehicle with respect to a geographical location in the environment.
15. The method of claim 14 , further comprising:
determining, using the processing circuit, a geographical location of the detected first target object based on the three-dimensional pose of the vehicle, wherein projecting the LiDAR data points onto the image data includes fusing the LiDAR data points with image data that is obtained for the environment around the vehicle.
16. The method of claim 11 , wherein automatically annotating the detected first target object present in the map data includes indicating a location of the detected first target object in the map data, and
the location of the detected first target object in the map data is based on a geographical location associated with the extracted subset of the LiDAR data points.
17. The method of claim 11 , further comprising:
determining, using the processing circuit, a three-dimensional pose of the detected first target object, wherein registering the detected first target object in the map data includes representing the detected first target object in the map data having the three-dimensional pose.
18. The method of claim 11 , wherein the registration criteria includes a criterion that is satisfied when the extracted subset of the LiDAR data points that corresponds to the detected first target object is associated with a distance that is less than a distance threshold.
19. A non-transitory computer-readable storage media storing instructions which, when executed by one or more computer processors, cause performance of operations comprising:
obtaining map data used for navigating a vehicle, wherein the map data identifies a plurality of objects in an environment of the vehicle;
obtaining image data of the image of the environment of the vehicle, wherein the image comprises one or more objects;
obtaining, using a LiDAR, LiDAR data points corresponding to the one or more objects;
projecting, using a processing circuit, the LiDAR data points onto the image data to generate fused data, the fused data comprising a LiDAR data portion and an image data portion;
detecting, using the processing circuit, a first target object and a second target object in the fused data;
extracting, using the processing circuit, a first subset of the LiDAR data points that corresponds to the detected first target object;
extracting, using the processing circuit, a second subset of the LiDAR data points that corresponds to the detected second target object;
automatically annotating, using the processing circuit, the map data, wherein the automatically annotating comprises:
in accordance with a determination that the extracted first subset of the LiDAR data points that corresponds to the detected first target object satisfies registration criteria associated with the detected first target object, automatically annotating the map data corresponding to the detected first target object, and
in accordance with a determination that the extracted second subset of the LiDAR data points that corresponds to the detected second target object does not satisfy the registration criteria associated with the detected second target object, not annotating the map data corresponding to the detected second target object; and
navigating, using a control circuit, the vehicle in the environment according to the annotated map data.