False positive object removal with surfel maps
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for using a surfel map to predict reflections in an environment. One of the methods includes receiving a surfel map comprising a plurality of surfels, wherein each surfel corresponds to a respective different location in an environment. Sensor data for one or more locations in the environment is obtained. The sensor data has been captured by one or more sensors of a vehicle. A range map that represents a projection of the surfel map is generated. Range data in the sensor data is compared to the range map to identify one or more locations in the range map that do not match the range data in the sensor data. The one or more locations in the range map that do not match the range data in the sensor data is classified as reflections.
1. A computer-implemented method comprising:
receiving a surfel map comprising a plurality of surfels, wherein each surfel corresponds to a respective different location on a surface of an object in an environment;
obtaining sensor data for one or more locations in the environment, the sensor data having been captured by one or more sensors of a vehicle;
generating a surface range map that represents a projection of the surfel map, wherein each pixel in the surface range map represents a distance from a location on a surface of an object to an observation point in the environment;
comparing range data in the sensor data to the surface range map generated from the surfel map to identify one or more locations in the surface range map that do not match the range data in the sensor data;
generating a classification result indicating that the one or more locations in the surface range map that do not match the range data in the sensor data correspond to reflections; and
providing the classification result indicating that the one or more locations correspond to the reflections to a planning subsystem for making autonomous driving decisions.
2. The method of claim 1 , wherein comparing the range data in the sensor data to the surface range map to identify the one or more locations in the surface range map that do not match the range data in the sensor data comprises:
obtaining an object mask of an object captured in the sensor data, wherein the object mask corresponds to range data of the object;
projecting the object mask onto the surface range map; and
comparing the range data of the object to the surface range map at a location indicated by the object mask of the object.
3. The method of claim 1 , wherein generating the classification result indicating that the one or more locations in the surface range map that do not match the range data in the sensor data correspond to the reflections comprises:
determining that, at the one or more locations, the range data in the sensor data is further than the surface range map generated from the surfel map; and
in response to determining that, at the one or more locations, the range data in the sensor data is further than the surface range map generated from the surfel map, classifying the one or more locations in the surface range map as the reflections.
4. The method of claim 1 , wherein the sensor data comprises lidar data captured by one or more lidar sensors of the vehicle.
5. The method of claim 1 , further comprising:
obtaining a likelihood score of an object detected in the sensor data at the one or more locations, wherein the likelihood score indicates a likelihood that the object is a vehicle type;
determining that the likelihood is less than a threshold value; and
in response to determining that the likelihood is less than the threshold value, determining, based at least in part on the classification result, that the object detected at the one or more locations is a false positive object caused by the reflections.
6. The method of claim 1 , wherein each surfel of the plurality of surfels is associated with a semantic label that represents an object type of a building.
7. The method of claim 1 , wherein the surface range map is an image.
8. The method of claim 1 , wherein the surface range map is continually updated while the vehicle drives in the environment.
9. The method of claim 1 , wherein the surface range map is continually updated multiple times per second.
10. The method of claim 1 , wherein the surface range map is continually generated by a parallel processing hardware.
11. A system comprising one or more computers and one or more storage devices storing instructions that are operable, when executed by the one or more computers, to cause the one or more computers to perform operations comprising:
receiving a surfel map comprising a plurality of surfels, wherein each surfel corresponds to a respective different location on a surface of an object in an environment;
obtaining sensor data for one or more locations in the environment, the sensor data having been captured by one or more sensors of a vehicle;
generating a surface range map that represents a projection of the surfel map, wherein each pixel in the surface range map represents a distance from a location on a surface of an object to an observation point in the environment;
comparing range data in the sensor data to the surface range map generated from the surfel map to identify one or more locations in the surface range map that do not match the range data in the sensor data;
generating a classification result indicating that the one or more locations in the surface range map that do not match the range data in the sensor data correspond to reflections; and
providing the classification result indicating that the one or more locations correspond to the reflections to a planning subsystem for making autonomous driving decisions.
12. The system of claim 11 , wherein comparing the range data in the sensor data to the surface range map to identify the one or more locations in the surface range map that do not match the range data in the sensor data comprises:
obtaining an object mask of an object captured in the sensor data, wherein the object mask corresponds to range data of the object;
projecting the object mask onto the surface range map; and
comparing the range data of the object to the surface range map at a location indicated by the object mask of the object.
13. The system of claim 11 , wherein generating the classification result indicating that the one or more locations in the surface range map that do not match the range data in the sensor data correspond to the reflections comprises:
determining that, at the one or more locations, the range data in the sensor data is further than the surface range map generated from the surfel map; and
in response to determining that, at the one or more locations, the range data in the sensor data is further than the surface range map generated from the surfel map, classifying the one or more locations in the surface range map as the reflections.
14. The system of claim 11 , wherein the sensor data comprises lidar data captured by one or more lidar sensors of the vehicle.
15. The system of claim 11 , wherein the operations further comprise:
obtaining a likelihood score of an object detected in the sensor data at the one or more locations, wherein the likelihood score indicates a likelihood that the object is a vehicle type;
determining that the likelihood is less than a threshold value; and
in response to determining that the likelihood is less than the threshold value, determining, based at least in part on the classification result, that the object detected at the one or more locations is a false positive object caused by the reflections.
16. The system of claim 11 , wherein each surfel of the plurality of surfels is associated with a semantic label that represents an object type of a building.
17. The system of claim 11 , wherein the surface range map is an image.
18. The system of claim 11 , wherein the surface range map is continually updated while the vehicle drives in the environment.
19. The system of claim 11 , wherein the surface range map is continually updated multiple times per second.
20. One or more non-transitory computer storage media encoded with computer program instructions that when executed by a plurality of computers cause the plurality of computers to perform operations comprising:
receiving a surfel map comprising a plurality of surfels, wherein each surfel corresponds to a respective different location on a surface of an object in an environment;
obtaining sensor data for one or more locations in the environment, the sensor data having been captured by one or more sensors of a vehicle;
generating a surface range map that represents a projection of the surfel map, wherein each pixel in the surface range map represents a distance from a location on a surface of an object to an observation point in the environment;
comparing range data in the sensor data to the surface range map generated from the surfel map to identify one or more locations in the surface range map that do not match the range data in the sensor data;
generating a classification result indicating that the one or more locations in the surface range map that do not match the range data in the sensor data correspond to reflections; and
providing the classification result indicating that the one or more locations correspond to the reflections to a planning subsystem for making autonomous driving decisions.