Method and apparatus for estimating pose of device

1. A method of estimating a pose of a device, the method comprising:

identifying a similar key frame that is similar to a current frame collected by the device from a key frame set, through an image feature extraction model that is trained by obtaining a global feature and a first set of local features from an original training image, obtaining a second set of local features from a modified training image that is acquired by modifying the original training image, creating a local feature pair by mapping the first set of local features to the second set of local features through self-supervised learning, and updating parameters of the image feature extraction model based on the global feature, the first set of local features, and the local feature pair;

obtaining data-related information between image frames based on a feature matching relationship between the current frame and the similar key frame; and

obtaining the pose of the device based on the data-related information.

2. The method of claim 1 , wherein the identifying of the similar key frame comprises:

determining a first similarity between each key frame in the key frame set and the current frame based on the global feature of the current frame;

determining a second similarity between each key frame in the key frame set and the current frame based on the first set of local features of the current frame; and

identifying the similar key frame based on the first similarity and the second similarity.

3. The method of claim 1 , wherein the obtaining of the data-related information comprises:

performing inter-frame feature matching on the current frame to obtain first data-related information between the image frames; and

updating the first data-related information based on the feature matching relationship between the current frame and the similar key frame to obtain second data-related information between the image frames.

4. The method of claim 3 , wherein the first data-related information includes a matching relationship between features of the image frames, and

wherein the updating the first data-related information comprises:

determining whether the features of the image frames are same features based on the feature matching relationship between the current frame and the similar key frame;

combining the same features based on a result of the determination; and

updating the first data-related information to obtain the second data-related information, based on the combined same features.

5. The method of claim 1 , further comprising:

obtaining the first set of local features and the global feature of the original training image through the image feature extraction model;

performing a conjugate rotation flip on the second set of local features of the modified image to obtain a conjugate rotation-flipped local feature of the modified training image;

obtaining the local feature pair based on the first set of local features of the original training image and the conjugate rotation-flipped local feature of the modified training image; and

updating the parameters of the image feature extraction model based on the local feature pair and the global feature.

6. The method of claim 1 : wherein the image feature extraction model comprises a first image feature extraction model and a second image feature extraction model, and

wherein identifying the similar key frame comprises:

obtaining the first set of local features and the global feature by inputting the original training image into the first image feature extraction model; and

obtaining the local feature pair by inputting the modified training image into the second image feature extraction model and pairing the first set of local features with the second set of local features obtained from the second image feature extraction model.

7. An apparatus for estimating a pose of a device, the apparatus comprising:

a memory storing computer readable instructions; and

at least one processor configured to execute the computer readable instructions to:

identify a similar key frame that is similar to a current frame collected by the device from a key frame set, through an image feature extraction model that is trained by obtaining a global feature and a first set of local features from an original training image, obtaining a second set of local features from a modified training image that is acquired by modifying the original training image, creating a local feature pair by mapping the first set of local features to the second set of local features through self-supervised learning, and updating parameters of the image feature extraction model based on the global feature, the first set of local features, and the local feature pair;

obtain data-related information between image frames based on a feature matching relationship between the current frame and the similar key frame; and

obtain the pose of the device based on the data-related information.

8. The apparatus of claim 7 , wherein the at least one processor is further configured to:

determine a first similarity between each key frame in the key frame set and the current frame based on the global feature of the current frame;

determine a second similarity between each key frame in the key frame set and the current frame based on the first set of local features of the current frame; and

identify the similar key frame based on the first similarity and the second similarity.

9. The apparatus of claim 7 , wherein the at least one processor is further configured to:

perform inter-frame feature matching on the current frame to obtain first data-related information between the image frames; and

update the first data-related information based on the feature matching relationship between the current frame and the similar key frame to obtain second data-related information between the image frames.

10. The apparatus of claim 9 , wherein the first data-related information includes a matching relationship between features of the image frames, and

wherein the at least one processor is further configured to:

determine whether the features of the image frames are same features based on the feature matching relationship between the current frame and the similar key frame;

combine the same features based on a result of the determination; and

update the first data-related information to obtain the second data-related information, based on the combined same features.

11. The apparatus of claim 7 , wherein the at least one processor is further configured to:

obtain the first set of local features and the global feature of the original training image through the image feature extraction model;

perform a conjugate rotation flip on the second set of local features of the modified training image to obtain a conjugate rotation-flipped local feature of the modified training image;

obtain the local feature pair based on the first set of local features of the original training image and the conjugate rotation-flipped local feature of the modified training image; and

update the parameters of the image feature extraction model based on the local feature pair and the global feature.

12. A non-transitory computer-readable storage medium storing a program that is executable by at least one processor to perform a method of estimating a pose of a device, the method comprising:

identifying a similar key frame that is similar to a current frame collected by the device from a key frame set, through an image feature extraction model that is trained by obtaining a global feature and a first set of local features from an original training image, obtaining a second set of local features from a modified training image that is obtained by modifying the original training image, creating a local feature pair by mapping the first set of local features to the second set of local features through self-supervised learning, and updating parameters of the image feature extraction model based on the global feature, the first set of local features, and the local feature pair;

obtaining data-related information between image frames based on a feature matching relationship between the current frame and the similar key frame; and

obtaining the pose of the device based on the data-related information.

13. The non-transitory computer-readable storage medium of claim 12 , wherein the identifying of the similar key frame comprises:

determining a first similarity between each key frame in the key frame set and the current frame based on the global feature of the current frame;

determining a second similarity between each key frame in the key frame set and the current frame based on the first set of local features of the current frame; and

identifying the similar key frame based on the first similarity and the second similarity.

14. The non-transitory computer-readable storage medium of claim 12 , wherein the obtaining of the data-related information comprises:

performing inter-frame feature matching on the current frame to obtain first data-related information between the image frames; and

updating the first data-related information based on the feature matching relationship between the current frame and the similar key frame to obtain second data-related information between the image frames.

15. The non-transitory computer-readable storage medium of claim 14 , wherein the first data-related information includes a matching relationship between features of the image frames, and

wherein the updating the first data-related information comprises:

determining whether the features of the image frames are same features based on the feature matching relationship between the current frame and the similar key frame;

combining the same features based on a result of the determination; and

updating the first data-related information to obtain the second data-related information, based on the combined same features.

16. The non-transitory computer-readable storage medium of claim 15 , wherein the method further comprises:

obtaining the first set of local features and the global feature of the original training image through the image feature extraction model;

performing a conjugate rotation flip on the second set of local features of the modified training image to obtain a conjugate rotation-flipped local feature of the modified training image;

obtaining the local feature pair based on the first set of local features of the original training image and the conjugate rotation-flipped local feature of the modified training image; and

updating the parameters of the image feature extraction model based on the local feature pair and the global feature.