Error estimation in real-time visual odometry system

1. A motion determination system in a machine, comprising:

one or more memories storing instructions; and

a controller configured to execute the instructions to perform operations including:

receiving, from a camera, a plurality of camera images representing an environment surrounding the machine;

receiving, from a range sensor, range data representing the environment;

identifying feature points in the plurality of camera images;

associating the range data with the identified feature points to generate one or more visual-odometry outputs;

determining a range matching error by computing an average error in a number of pixels between the feature points in the camera images and points in the camera images corresponding to the range data associated with the feature points;

determining a plurality of figure of merits, wherein each of the plurality of figure of merits is associated with one of a plurality of parameters and is indicative of an accuracy of the visual-odometry outputs, the plurality of parameters affecting the generation of the one or more visual-odometry outputs and including at least the range matching error;

calculating a combined figure of merit based on the plurality of figure of merits; and

calculating an error estimate for the one or more visual-odometry outputs based on the combined figure of merit.

2. The system of claim 1 , wherein the one or more visual-odometry outputs include a translation velocity and a rotation velocity.

3. The system of claim 1 , wherein the plurality of parameters further include a number of matching feature points, a percentage of inlier feature points, an inlier error, and a geometric distribution of the inlier feature points.

4. The system of claim 1 , wherein the combined figure of merit is calculated by multiplying each of the plurality of figure of merits.

5. The system of claim 1 , wherein calculating the error estimate includes:

obtaining a target performance-error estimate for the one or more visual-odometry outputs; and

calculating the error estimate based on the obtained target performance-error estimate and the combined figure of merit.

6. The system of claim 5 , wherein the error estimate is calculated by dividing the obtained target performance-error estimate with the combined figure of merit.

7. The system of claim 3 , wherein

the number of matching feature points is a number of common feature points between the plurality of camera images.

8. A computer-implemented method for determining motion of a machine, the method comprising:

receiving, from a camera, a plurality of camera images representing an environment surrounding the machine;

receiving, from a range sensor, range data representing the environment;

identifying feature points in the plurality of camera images;

associating the range data with the identified feature points to generate one or more visual-odometry outputs;

determining a range matching error by computing an average error in a number of pixels between the feature points in the camera images and points in the camera images corresponding to the range data associated with the feature points;

determining a plurality of figure of merits, wherein each of the plurality of figure of merits is associated with one of a plurality of parameters and is indicative of an accuracy of the visual-odometry outputs, the plurality of parameters affecting the generation of the one or more visual-odometry outputs and including at least the range matching error;

calculating a combined figure of merit based on the plurality of figure of merits; and

calculating an error estimate for the one or more visual-odometry outputs based on the combined figure of merit.

9. The method of claim 8 , wherein the one or more visual-odometry outputs include a translation velocity and a rotation velocity.

10. The method of claim 8 , wherein the plurality of parameters include a number of matching feature points, a percentage of inlier feature points, an inlier error, and a geometric distribution of the inlier feature points.

11. The method of claim 8 , wherein the combined figure of merit is calculated by multiplying each of the plurality of figure of merits.

12. The method of claim 8 , wherein calculating the error estimate includes:

obtaining a target performance-error estimate for the one or more visual-odometry outputs; and

calculating the error estimate based on the obtained target performance-error estimate and the combined figure of merit.

13. The method of claim 12 , wherein the error estimate is calculated by dividing the obtained target performance-error estimate with the combined figure of merit.

14. The method of claim 10 , wherein

the number of matching feature points is a number of common feature points between the plurality of camera images.

15. A machine comprising:

a range sensor that generates range data representing an environment surrounding the machine;

a camera that generates camera images representing the environment; and

a controller including:

one or more memories storing instructions; and

a processor configured to execute the instructions to process the range data and the camera images to perform operations including:

receiving, from a camera, a plurality of camera images representing an environment surrounding the machine;

receiving, from a range sensor, range data representing the environment;

identifying feature points in the plurality of camera images;

associating the range data with the identified feature points to generate one or more visual-odometry outputs;

determining a range matching error by computing an average error in a number of pixels between the feature points in the camera images and points in the camera images corresponding to the range data associated with the feature points;

determining a plurality of figure of merits, wherein each of the plurality of figure of merits is associated with one of a plurality of parameters and is indicative of an accuracy of the visual-odometry outputs, the plurality of parameters affecting the generation of the one or more visual-odometry outputs and including at least the range matching error;

calculating a combined figure of merit based on the plurality of figure of merits; and

calculating an error estimate for the one or more visual-odometry outputs based on the combined figure of merit.

16. The machine of claim 15 , wherein the one or more visual-odometry outputs include a translation velocity and a rotation velocity.

17. The machine of claim 15 , wherein the plurality of parameters include a number of matching feature points, a percentage of inlier feature points, an inlier error, and a geometric distribution of the inlier feature points.

18. The machine of claim 15 , wherein the combined figure of merit is calculated by multiplying each of the plurality of figure of merits.

19. The machine of claim 15 , wherein calculating the error estimate includes:

obtaining a target performance-error estimate for the one or more visual-odometry outputs; and

calculating the error estimate based on the obtained target performance-error estimate and the combined figure of merit.

20. The machine of claim 19 , wherein the error estimate is calculated by dividing the obtained target performance-error estimate with the combined figure of merit.