Method for 3D reconstruction from satellite imagery
The present disclosure relates to a method for 3D reconstruction from satellite imagery using deep learning, said method comprising providing ( 101 ) at least two overlapping 2D satellite images, providing ( 102 ) imaging device parameters for the at least two overlapping 2D satellite images, providing ( 103 ) at least one trained Machine Learning Network, MLN, able to predict depth maps, said trained MLN being trained on a training set comprising multi-view geocoded 3D ground truth data and predicting ( 104 ) a depth map of the at provided at least two 2D satellite images using the trained at least one MLN and based on the corresponding imaging device parameters.
1. A method for 3D reconstruction from satellite imagery, the method comprising:
providing (a) at least two at least partially overlapping 2D satellite images and (b) imaging device parameters for the at least two partially overlapping 2D satellite images to at least one trained Machine Learning Network (MLN), wherein the MLN (a) has been trained on a training set comprising multi-view 3D geocoded ground truth data and (b) is configured to compute a depth map of the at least two 2D satellite images based on the imaging device parameters;
receiving the computed depth map; and
generating a textured geocoded 3D surface model based on the computed depth map.
2. The method according to claim 1 , wherein the textured geocoded 3D surface model is represented as a mesh.
3. The method according to claim 2 , wherein the mesh comprises a plurality of nodes interconnected by one or more edges, and wherein surfaces are defined by the edges of the mesh.
4. The method according to claim 3 , wherein each of the plurality of nodes is each associated to a 3D coordinate of a geographical coordinate system.
5. The method according to claim 3 , wherein each of the surfaces is associated to texture information.
6. The method according to claim 1 , wherein the textured geocoded 3D surface model is represented as a surface representation.
7. The method according to claim 1 , wherein the textured geocoded 3D surface model is represented as a voxel representation.
8. A method for 3D reconstruction from satellite imagery, the method comprising:
providing (a) at least two at least partially overlapping 2D satellite images and (b) imaging device parameters for the at least two partially overlapping 2D satellite images to at least one trained Machine Learning Network (MLN),
wherein the MLN:
(a) has been trained on a training set comprising multi-view 3D geocoded ground truth data,
(b) the multi-view 3D geocoded ground truth data is at least one of (i) geocoded and extracted or (ii) rendered from a textured geocoded 3D surface model, and
(c) is configured to compute a depth map of the at least two partially overlapping 2D satellite images based on the corresponding imaging device parameters, wherein the computing of the depth map comprises associating data of the depth map to geocoded coordinate data.
9. The method according to claim 8 , wherein the textured geocoded 3D surface model is represented as a mesh.
10. The method according to claim 9 , wherein the mesh comprises a plurality of nodes interconnected by one or more edges, and wherein surfaces are defined by the edges of the mesh.
11. The method according to claim 10 , wherein each of the plurality of nodes is each associated to a 3D coordinate of a geographical coordinate system.
12. The method according to claim 10 , wherein each of the surfaces is associated to texture information.
13. The method according to claim 8 , wherein the textured geocoded 3D surface model is represented as a surface representation.
14. The method according to claim 8 , wherein the textured geocoded 3D surface model is represented as a voxel representation.
15. The method according to claim 8 , wherein data of the training set has a higher resolution than a resolution of the at least two partially overlapping 2D satellite images.
16. The method according to claim 8 , wherein the multi-view 3D geocoded ground truth data is real world data.
17. The method according to claim 8 , wherein the multi-view 3D geocoded ground truth data is extracted from a textured geocoded 3D surface model, the geocoded 3D surface model being based on real world data.
18. The method according to claim 8 , wherein the multi-view 3D geocoded ground truth data comprises measurement data of at least one of images, LIDAR, radar, or sonar.
19. The method according to claim 8 , wherein the multi-view 3D geocoded ground truth data (a) relates to a plurality of geographical areas and (b) comprises a plurality of reference images and/or measurement data representing each of the plurality of geographical areas from different angles.
20. A method for training a Machine Learning Network (MLN) for 3D reconstruction from satellite imagery, the method comprising:
providing at least one trained MLN configured to compute depth maps,
training the MLN on a training set comprising at least one textured geocoded 3D surface model for training purposes, and
extracting or rendering multi-view 3D geocoded ground truth data from the textured geocoded 3D surface model for training purposes.
21. The method according to claim 20 , wherein the textured geocoded 3D surface model is represented as a mesh.
22. The method according to claim 21 , wherein the mesh comprises a plurality of nodes interconnected by means of edges, wherein surfaces are defined by the edges of the mesh.
23. The method according to claim 22 , wherein each of the plurality of nodes is each associated to a 3D coordinate of a geographical coordinate system.
24. The method according to claim 22 , wherein each of the surfaces are associated to texture information.
25. The method according to claim 22 , wherein the textured generated geocoded 3D surface model is represented as a voxel representation.