Method and apparatus with pose estimation

1. A processor-implemented method with pose estimation, the method comprising:

tracking a position of a feature point extracted from image information comprising a plurality of image frames, the image information being received from an image sensor;

predicting a current state variable of an estimation model for determining a pose of an electronic device, based on motion information received from a motion sensor and an error covariance of a previous state variable and the current state variable;

determining noise due to an uncertainty of the estimation model based on a residual between a first position of the feature point extracted from the image frames and a second position of the feature point predicted based on the current state variable;

updating the current state variable based on the current state variable, the tracked position of the feature point, and the noise; and

determining the pose of the electronic device based on the updated current state variable.

2. The method of claim 1 , wherein the determining of the noise comprises:

determining the residual between the first position of the feature point extracted from a current image frame among the image frames and the second position of the feature point predicted based on the current state variable; and

determining the noise due to the uncertainty of the estimation model based on the residual.

3. The method of claim 2 , wherein the determining of the noise comprises determining the noise by expressing, as a sum of a ground-truth and an error, the residual between the first position of the feature point extracted from the current image frame and the second position of the feature point predicted based on the current state variable.

4. The method of claim 3 , wherein the determining of the noise comprises determining the noise by approximating a result of expressing the residual as the sum of the ground-truth and the error, using a tuning parameter of the estimation model.

5. The method of claim 1 , wherein the uncertainty of the estimation model is determined using an estimation covariance according to a simultaneous localization and mapping (SLAM) technique based on an extended Kalman filter.

6. The method of claim 1 , wherein the predicting of the current state variable comprises predicting, based on the motion information and a previous state variable corresponding to a previous image frame among the image frames, the error covariance of the previous state variable and the current state variable corresponding to a current image frame among the image frames.

7. The method of claim 6 , wherein the updating of the current state variable comprises updating the current state variable based on the current state variable, the error covariance of the previous state variable, the tracked position of the feature point, and the noise.

8. The method of claim 1 , wherein the tracking of the position of the feature point comprises:

extracting a second feature point included in a current image frame among the image frames; and

tracking a position of the second feature point through matching between a first feature point extracted from a previous image frame before the current image frame among the image frames and the second feature point.

9. The method of claim 1 , wherein the estimation model comprises an estimation model using a filter-based visual-inertial odometry (VIO).

10. The method of claim 1 , wherein the electronic device comprises either one or both of:

a wearable device comprising a smart glass; and

a head-mounted device (HMD) comprising any one or any combination of any two or more of an augmented reality (AR) device, a virtual reality (VR) device, and a mixed reality (MR) device.

11. A non-transitory computer-readable storage medium storing instructions that, when executed by one or more processors, configure the one or more processors to perform the method of claim 1 .

12. An apparatus with pose estimation, the apparatus comprising:

an image sensor configured to sense image information comprising a plurality of image frames;

a motion sensor configured to sense motion information of an electronic device; and

one or more processors configured to:

track a feature point by extracting the feature point from the image information;

predict a current state variable of an estimation model for determining a pose of the electronic device based on the motion information and an error covariance of a previous state variable and the current state variable;

determine noise due to an uncertainty of the estimation model based on a residual between a first position of the feature point extracted from the image frames and a second position of the feature point predicted based on the current state variable;

update the current state variable based on the current state variable, a position of the tracked feature point, and the noise; and

determine the pose of the electronic device based on the updated current state variable.

13. The apparatus of claim 12 , wherein, for the determining of the noise, the one or more processors are configured to:

determine the residual between the first position of the feature point extracted from a current image frame among the image frames and the second position of the feature point predicted based on the current state variable; and

determine the noise due to the uncertainty of the estimation model based on the residual.

14. The apparatus of claim 13 , wherein, for the determining of the noise, the one or more processors are configured to determine the noise by expressing, as a sum of a ground-truth and an error, the residual between the first position of the feature point extracted from the current image frame and the second position of the feature point predicted based on the current state variable.

15. The apparatus of claim 14 , wherein, for the determining of the noise, the one or more processors are configured to determine the noise by approximating a result of expressing the residual as the sum of the ground-truth and the error, using a tuning parameter of the estimation model.

16. The apparatus of claim 12 , wherein the uncertainty of the estimation model is determined using an estimation covariance according to a simultaneous localization and mapping (SLAM) technique based on an extended Kalman filter.

17. The apparatus of claim 12 , wherein, for the predicting of the current state variable, the one or more processors are configured to predict, based on the motion information and a previous state variable corresponding to a previous image frame among the image frames, the error covariance of the previous state variable and the current state variable corresponding to a current image frame among the image frames.

18. The apparatus of claim 17 , wherein, for the updating of the current state variable, the one or more processors are configured to update the current state variable based on the current state variable, the error covariance of the previous state variable, the position of the tracked feature point, and the noise.

19. The apparatus of claim 12 , wherein, for the tracking of the position of the feature point, the one or more processors are configured to:

extract a second feature point included in a current image frame among the image frames; and

track a position of the second feature point through matching between a first feature point extracted from a previous image frame before the current image frame among the image frames and the second feature point.

20. The apparatus of claim 12 , wherein the estimation model comprises an estimation model using a filter-based visual-inertial odometry (VIO).

21. The apparatus of claim 12 , wherein the electronic device comprises either one or both of:

a wearable device comprising a smart glass; and

a head-mounted device (HMD) comprising any one or any combination of any two or more of an augmented reality (AR) device, a virtual reality (VR) device, and a mixed reality (MR) device.

22. A wearable electronic device comprising:

an image sensor configured to sense image information comprising a plurality of image frames;

a motion sensor configured to sense motion information of the wearable electronic device;

one or more processors configured to:

track a feature point by extracting the feature point from the image information;

predict a current state variable of an estimation model for determining a pose of the wearable electronic device based on the motion information and an error covariance of a previous state variable and the current state variable;

determine noise due to an uncertainty of the estimation model based on a residual between a first position of the feature point extracted from the image frames and a second position of the feature point predicted based on the current state variable;

update the current state variable based on the current state variable, a position of the tracked feature point, and the noise;

determine the pose of the wearable electronic device based on the updated current state variable; and

generate virtual content based on the determined pose; and

a display configured to display the virtual content.

23. A processor-implemented method with pose estimation, the method comprising:

predicting a current state variable of an estimation model for determining a pose of an electronic device, based on an error covariance of a previous state variable and the current state variable;

determining noise due to an uncertainty of the estimation model based on a residual between a first position of a feature point extracted from image frames and a second position of the feature point predicted based on the current state variable;

updating the current state variable based on the noise; and

determining the pose of the electronic device based on the updated current state variable.

24. The method of claim 23 , wherein the current state variable comprises a state variable corresponding to a motion sensor and a state variable corresponding to an image sensor.

25. The method of claim 24 , wherein the state variable corresponding to the motion sensor comprises any one or any combination of any two or more of an orientation error of an image frame among the image frames, a bias error of the motion sensor, a velocity error, an acceleration bias error, and a position error.

26. The method of claim 24 , wherein the state variable corresponding to the image sensor comprises an error of a position of a feature point viewed at an image frame of the image frames.