Map refinement using feature extraction from images
The present disclosure relates to refining maps for vehicles, for example using feature extraction to identify parking areas and their contents from high resolution satellite or drone images. Such maps can be useful for autonomous valet parking (AVP) features of autonomous vehicles.
1. A system for refining maps for vehicles, the system comprising:
one or more processors in communication with a vehicle;
wherein the one or more processors are configured to:
receive location data from the vehicle;
retrieve map data associated with the location data, the map data including a previously-generated map;
retrieve image data corresponding to the location data, the image data including an image of a portion of Earth's surface;
perform feature extraction on the image data to identify one or more parking spots;
generate a refined map by revising the previously-generated map to include the one or more parking spots;
store the refined map as the map data associated with the location data; and
transmit the map data associated with the location data to the vehicle.
2. The system of claim 1 , the one or more processors further configured to transmit the map data associated with the location data to a plurality of vehicles.
3. The system of claim 1 , wherein identifying the one or more parking spots comprises a low-level parking spot identification followed by a grouping of parking spots to identify a parking area.
4. The system of claim 3 , wherein the low-level parking spot identification includes retrieving a set of parking spot templates and performing a normalized cross correlation over the image data using the parking spot templates.
5. The system of claim 4 , wherein the parking spot templates comprise a plurality of parking lot patches, each patch corresponding to a unique configuration of parking lines.
6. The system of claim 4 , wherein the grouping of parking spots includes identifying parking rows comprised of identified parking spots, and determining whether the identified parking rows fall within a distance threshold of one another.
7. The system of claim 3 , wherein identifying the one or more parking spots further comprises identifying physical boundaries of the parking area.
8. The system of claim 1 , wherein the image has a resolution of at least 30 cm.
9. The system of claim 1 , wherein the processors in the vehicle are further configured to perform autonomous valet parking based on the received map data.
10. A method of refining a map, comprising:
determining a location of a vehicle;
retrieving a map associated with the location, the map being previously-generated;
retrieving an image of a portion of Earth's surface corresponding to the location;
performing feature extraction on the image to identify one or more parking spots;
revising the previously-generated map to include the one or more parking spots;
storing the revised map as the map associated with the location; and
transmitting the map associated with the location to the vehicle.
11. The method of claim 10 , further comprising transmitting the map data associated with the location data to a plurality of vehicles.
12. The method of claim 10 , wherein identifying the one or more parking spots comprises a low-level parking spot identification followed by a grouping of parking spots to identify a parking area.
13. The method of claim 12 , wherein the low-level parking spot identification includes retrieving a set of parking spot templates and performing a normalized cross correlation over the image data using the parking spot templates.
14. The method of claim 13 , wherein the parking spot templates comprise a plurality of parking lot patches, each patch corresponding to a unique configuration of parking lines.
15. The method of claim 12 , wherein the grouping of parking spots includes identifying parking rows comprised of identified parking spots, and determining whether the identified parking rows fall within a distance threshold of one another.
16. The method of claim 12 , wherein identifying the one or more parking spots further comprises identifying physical boundaries of the parking area.
17. The method of claim 10 , wherein the image has a resolution of at least 30 cm.
18. The method of claim 10 , further comprising performing autonomous valet parking based on the received map data.
19. A non-transitory computer-readable medium including instructions, which when executed by one or more processors, cause the one or more processors to perform a method comprising:
determining a location of a vehicle;
retrieving a map associated with the location, the map being previously-generated;
retrieving an image of a portion of Earth's surface corresponding to the location;
performing feature extraction on the image to identify one or more parking spots;
revising the previously-generated map to include the one or more parking spots;
storing the revised map as the map associated with the location; and
transmitting the map associated with the location to the vehicle.