Learnable localization using images

1. An apparatus for generating a model for pose estimation of a system, the apparatus comprising a processor circuitry, the processor circuitry being configured to:

obtain training data for a plurality of locations, the training data comprising a number of captured images captured by an image capturing device, and a respective number of poses of the captured images, wherein a respective pose, of the respective poses of the image capturing device, associated with a respective captured image, of the captured images, comprises a respective location, of the locations, and an orientation of the image capturing device during capture of the captured image;

generate from the training data at least one data sample for each of the captured images, wherein the at least one data sample for the respective captured image comprises an assignment of the respective captured image and at least one other image selected from the training data to the corresponding respective poses;

train a neural network with a data set comprising the at least one data sample to estimate a respective pose of a localization image from:

the localization image;

at least one additional image from the training data;

the respective pose of the at least one additional image; and

calculate an estimation accuracy measure by comparing a pose change between a plurality of estimated poses for the localization image to at least one pose change between reference images used for calculating the plurality of estimated poses.

2. The apparatus according to claim 1 , wherein the generating the at least one data sample comprises:

selecting at least one reference image from the training data;

selecting at least one proximate image having a pose proximate to a pose of the at least one reference image from the training data; and

assigning the at least one reference image and the at least one proximate image to respective poses of each proximate image relative to each reference image.

3. The apparatus according to claim 2 , wherein the at least one reference image comprises a single reference image and the assigning designates the respective poses of each proximate image relative to the single reference image.

4. The apparatus according to claim 2 , wherein the at least one reference image comprises a plurality of reference images and the assigning designates respective poses of each proximate image relative to each of the reference images.

5. The apparatus according to claim 2 , wherein respective poses of the at least one proximate image are within a specified distance from a pose of at least one of the reference images.

6. The apparatus according to claim 1 , wherein the processor circuitry is further configured to introduce an error into the training data by changing at least one pose in the data samples prior to the training.

7. The apparatus according to claim 1 , wherein the processor circuitry is further configured to track errors in estimation of the poses during the training, and to calculate respective confidence values for poses estimated by the neural network based on the tracked errors.

8. The apparatus according to claim 1 , wherein at least one of the data samples further comprises a semantic segmentation computed for at least one image mapped in the data set.

9. The apparatus according to claim 1 , wherein at least one of the data samples further comprises a depth map computed for at least one image mapped in the data set.

10. The apparatus according to claim 1 , wherein the neural network comprises a deep neural network (DNN).

11. A method for generating a neural network for pose estimation of a system, the method comprising:

obtaining training data for a plurality of locations, the training data comprising a number of captured images captured by an image capturing device, and a respective number of poses of the captured images, wherein a respective pose, of the respective poses of the image capturing device, associated with a respective captured image, of the captured images, comprises a respective location, of the locations, and an orientation of the image capturing device during capture of the captured image;

generating from the training data at least one data sample for each of the captured images, wherein the at least one data sample for the respective captured image comprises an assignment of said the respective captured image and at least one other image selected from the training data to the corresponding respective poses;

training a neural network with a data set comprising the at least one data sample to estimate a respective pose of a localization image from:

the localization image;

at least one additional image from the training data; and

a respective pose of each of the at least one additional image; and

calculating an estimation accuracy measure by comparing a pose change between a plurality of estimated poses for the localization image to at least one pose change between reference images used for calculating the plurality of estimated poses.

12. The method according to claim 11 , wherein, for at least one of the data samples, the assignment comprises:

selecting a single reference image from the training data;

selecting at least one proximate image selected training data based on proximity to the reference image; and

designating respective poses of the at least one proximate image relative to the single reference image.

13. The method according to claim 11 , wherein, for at least one of the data samples, the assignment comprises:

selecting a plurality of reference images selected from the training data;

selecting at least one proximate image from the training data based on proximity to the reference images; and

designating respective poses of each proximate image relative to each reference image.

14. The method according to claim 11 , further comprising introducing an error into the training data by changing at least one pose in the data samples prior to the training.

15. The method according to claim 11 , further comprising tracking errors in estimation of the poses during the training, and calculating a respective confidence value for poses estimated by the neural network based on the tracking.

16. An apparatus for image-based localization of a system in an area of interest, comprising:

a memory configured to store images captured by an image capturing device of the system at a plurality of locations in the area of interest, and respective poses of the image capturing device associated with the stored images at each of the locations, wherein a pose, of the respective poses, of a respective image, of the stored images, comprises a location, of the locations, and an orientation of the capturing device associated with the respective image; and

a processor circuitry configured to:

obtain a localization image captured by the image capturing device;

retrieve from the memory a subset of the stored images and the respective poses;

provide to a trained neural network:

the localization image;

the retrieved images; and

the respective poses of the retrieved images; and

obtain from the trained neural network at least one estimated pose for the localization image; and

calculate an estimation accuracy measure by comparing a pose change between a plurality of estimated poses for the localization image to at least one pose change between reference images used for calculating the plurality of estimated poses.

17. The apparatus according to claim 16 , wherein the processor circuitry is further configured to:

determine an approximate location of the image capturing device during the capturing of the localization image; and

select the subset of stored images based on the approximate location.

18. The apparatus according to claim 16 , wherein the neural network outputs a single estimated pose relative to a pose of a single reference image selected from the retrieved images.

19. The apparatus according to claim 16 , wherein the neural network outputs the plurality of estimated poses, each of the estimated poses being relative to a pose of a different reference image selected from the retrieved images, and wherein the processor circuitry is further configured to calculate a final estimated pose from the plurality of estimated poses.

20. The apparatus according to claim 16 , wherein the trained neural network comprises a deep neural network (DNN).

21. The apparatus according to claim 16 , wherein the trained neural network is further provided with a semantic segmentation computed for at least one of the images provided to the trained neural network.

22. The apparatus according to claim 16 , wherein the trained neural network is further provided with a depth map computed for at least one of the images provided to the trained neural network.

23. A method for image-based localization of a system in an area of interest, comprising:

obtaining a localization image captured by an image capturing device of the system;

providing to a trained neural network:

the localization image;

a subset of images selected from a plurality of images captured during a mapping of the area of interest; and

respective poses associated with the subset of images, wherein a pose, of the poses, of an image, of the images, comprises a location and an orientation associated with the image capturing device when capturing the image;

obtaining from the trained neural network at least one estimate of a pose of the localization image; and

calculate an estimation accuracy measure by comparing a pose change between a plurality of estimated poses for the localization image to at least one pose change between reference images used for calculating the plurality of estimated poses.

24. The method according to claim 23 , wherein the subset of images is selected based on a proximity to an estimated location of the localization image.

25. The method according to claim 23 , wherein the neural network outputs a single estimated pose relative to a pose of a single reference image selected from the subset of images.

26. The method according to claim 25 , further comprising obtaining an approximate location of the image capturing device from a satellite navigation system, the reference image being selected based on the approximate location.

27. The method according to claim 23 , wherein the neural network outputs the plurality of estimated poses, each of the estimated poses being relative to a pose of a different reference image selected from the subset of images, further comprising calculating a final estimated pose from the plurality of estimated poses.

28. The method according to claim 23 , wherein the trained neural network comprises a deep neural network (DNN).