Method and device for camera pose estimation

1. A method comprising:

acquiring a plurality of training images through a capturing component;

acquiring a plurality of training camera poses of the capturing component corresponding to the training images through a pose sensor disposed corresponding to the capturing component; and

training a camera pose estimation model according to the training images and the training camera poses of the capturing component;

wherein the operation of training the camera pose estimation model comprises:

extracting training feature sets from the training images by using a feature extracting module;

calculating estimating camera poses according to the training feature sets by using a pose regressor; and

training the camera pose estimation model according to loss functions corresponding to differences between the estimating camera poses and the training camera poses, wherein the loss functions include a plurality of inputs, and wherein the inputs include the estimating camera poses, and the estimating camera poses are a function of the training feature sets.

2. The method as claimed in claim 1 , wherein one of the training images is captured while a corresponding one of the training camera poses is sensed.

3. The method as claimed in claim 1 , wherein the pose regressor comprises a plurality of trainable parameters, and training the camera pose estimation model according to the loss functions comprises:

determining the trainable parameters of the camera pose estimation model by minimizing the loss functions.

4. The method as claimed in claim 1 , wherein the feature extracting module comprises a plurality of first trainable parameters, and the pose regressor comprises a plurality of second trainable parameters, and training the camera pose estimation model according to the loss functions comprises:

determining the first trainable parameters of the feature extracting module and the second trainable parameters of the pose regressor by minimizing the loss functions.

5. The method as claimed in claim 1 further comprising:

acquiring training auxiliary information through one or more auxiliary sensors;

wherein the inputs of the loss functions include the estimating camera poses, and the estimating camera poses are a function of the training feature sets and the training auxiliary information.

6. The method as claimed in claim 5 , wherein the training auxiliary information comprises one or more of training angular velocity information and training acceleration information.

7. A method for estimating a camera pose of a capturing component, comprising:

providing a camera pose estimation model;

training the camera pose estimation model to obtain a trained camera pose estimation model;

providing the capturing component;

acquiring an input image and a reference image through the capturing component; and

estimating the camera pose of the capturing component by using the input image, reference image and the trained camera pose estimation model;

wherein the camera pose estimation model is trained according to training images and training camera poses, and wherein the training camera poses are acquired through a pose sensor.

8. The method as claimed in claim 7 , wherein the operation of estimating the camera pose of the capturing component comprises:

extracting a first feature set from the input image and a second feature set from the reference image; and

estimating the camera pose of the capturing component according to the first feature set and the second feature set by using a pose regressor.

9. The method as claimed in claim 8 , wherein the pose regressor is trained according to the training camera poses and training feature sets extracted from the training images.

10. The method as claimed in claim 7 further comprising:

acquiring auxiliary information through one or more auxiliary sensor;

wherein the operation of estimating the camera pose of the capturing component comprises:

estimating the camera pose of the capturing component by using the input image, the reference image, the auxiliary information, and trained camera pose estimation model.

11. The method as claimed in claim 10 , wherein the auxiliary information comprises one or more of angular velocity information and acceleration information.

12. The method as claimed in claim 7 , wherein the camera pose estimation is trained by calculating estimating camera poses according to the training images with the camera pose estimation model and determining trainable parameters of the camera pose estimation model by minimizing loss functions corresponding to differences between the estimating camera poses and the training camera poses.

13. A device for estimating a camera pose of a capturing component, comprising:

the capturing component;

one or more processing components electrically connected to the capturing component;

a memory electrically connected to the one or more processing components; and

one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processing components, the one or more programs comprising instructions for:

providing a camera pose estimation model;

training the camera pose estimation model to obtain a trained camera pose estimation model;

acquiring an input image and a reference image through the capturing component; and

estimating a camera pose of the capturing component by using the input image, the reference image, and the trained camera pose estimation model;

wherein the camera pose estimation model is trained according to training images and training camera poses, and wherein the training camera poses are acquired through a pose sensor.

14. The device as claimed in claim 13 , wherein the one or more programs further comprises instructions for:

extracting a first feature set from the input image and a second feature set from the reference image; and

estimating the camera pose of the capturing component according to the first feature set and the second feature set by using a pose regressor.

15. The device as claimed in claim 13 , wherein the pose regressor is trained according to the training camera poses and training feature sets extracted from the training images.

16. The device as claimed in claim 13 , wherein the one or more programs further comprises instructions for:

acquiring auxiliary information through one or more auxiliary sensor; and

estimating the camera pose of the capturing component by using the input image, the reference image, the auxiliary information, and the trained camera pose estimation model.

17. The device as claimed in claim 16 , wherein the auxiliary information comprises one or more of angular velocity information and acceleration information.

18. The device as claimed in claim 13 , wherein the camera pose estimation is trained by calculating estimating camera poses according to the training images with the camera pose estimation model and determining trainable parameters of the camera pose estimation model by minimizing loss functions corresponding to differences between the estimating camera poses and the training camera poses.