Granted Patent B2

US 12,259,259 · App. 18/020,595 · Granted Mar 25, 2025

Inventors: Ying Zhang (Singapore, SG); Jagannadan Varadarajan (Singapore, SG); Roger Zimmermann (Singapore, SG); Guanfeng Wang (Singapore, SG)

Assignee: GRABTAXI HOLDINGS PTE. LTD.

G01C21/3852G01C21/3848G06N3/0475G06N3/094

Abstract

Claims (27)

1. A method for generating map data comprising:

training a generator neural network by

acquiring training data elements for a generative adversarial network, each training data element comprising a satellite image of a geographical area, a road usage image of the geographical area which shows routes in the geographical area which may be used for driving a vehicle and a map data image of the geographical area, the satellite image an image of the geographical area and acquired from a satellite image database; and

training the generative adversarial network by training the generator neural network, using the training data elements, to generate, for a satellite image and a road usage image of a training data element, the map data image of the training data element; and

generating map data for a first geographical area by

acquiring first road usage information specifying which parts of the first geographical area have been used for driving a vehicle, the first road usage information including one or more raw global positioning system (GPS) traces received from a plurality of vehicles that have driven in the first geographical area, the one or more raw GPS traces provided in a textual form;

acquiring a first satellite image of the first geographical area;

forming a first road usage image of the first geographical area by converting the one or more raw GPS traces from the textual form into the first road usage image using GPS image rendering, the first road usage image having pixels, each pixel corresponding to a respective part of the first geographical area, such that each pixel has a pixel value indicating whether the part of the first geographical area, to which the pixel corresponds, is specified by the first road usage information to have been used for driving a vehicle;

feeding the first satellite image and the first road usage image to the trained generator neural network to produce a generated map data image of the first geographical area; and

outputting the generated map data image for the first geographical area.

2. The method of claim 1 , wherein in at least some of the training data elements, the satellite image shows routes which are not shown in the road usage image.

3. The method of claim 1 , wherein in at least some of the training data elements, the road usage image shows routes which are not shown in the satellite image.

4. The method of claim 1 , wherein the first map data image is a visual map of the first geographic area.

5. The method of claim 1 , wherein the first map data image specifies a road network of the first geographic area.

6. The method of claim 1 , wherein the generator neural network comprises a U-Net.

7. The method of claim 1 , wherein the generator neural network comprises an image concatenation layer for concatenating the satellite image and the road usage image and for concatenating the first satellite image and the first road usage image.

8. The method of claim 1 , wherein training the generative adversarial network comprises training a discriminator of the generative adversarial network to decide whether map data image fed to the discriminator is a real map data image for a satellite image fed to the discriminator or a fake map data image for the satellite image.

9. The method of claim 1 , wherein the generative adversarial network comprises a discriminator comprising a convolutional network.

10. The method of claim 9 , wherein the discriminator comprises an image concatenation layer for concatenating a satellite image and a map data image which is fed to the discriminator to have the discriminator decide whether it is a real map data image for the satellite image or a fake map data image for the satellite image.

11. The method of claim 1 , wherein forming the first road usage image comprises assigning a first pixel value of two predetermined pixel values to a pixel if the pixel corresponds to a part of the first geographic area which is specified by the first road usage information to have been used for driving a vehicle and assigning a second pixel value of two predetermined pixel values to a pixel if the pixel corresponds to a part of the first geographic area which is not specified by the first road usage information to have been used for driving a vehicle.

12. The method of claim 1 , wherein the first road usage information is positioning system traces of a plurality of vehicles.

13. The method of claim 1 , wherein acquiring first road usage information comprises acquiring positioning system traces of a plurality of vehicles of a transport service.

14. The method of claim 1 , further comprising navigating one or more vehicles by the generated map data image.

15. The method of claim 1 , further comprising transmitting the generated map data image to a vehicle and controlling the vehicle using the generated map data image.

16. A server computer comprising a radio interface, a memory interface and a processing unit configured to perform the method of claim 1 .

17. A computer program element comprising program instructions, which, when executed by one or more processors, cause the one or more processors to perform the method of claim 1 .

18. A computer-readable medium comprising program instructions, which, when executed by one or more processors, cause the one or more processors to perform the method of claim 1 .

Assignments (3)

Priority Claims (1)

Continuity (1)

References Cited (18)