Geo-registering an aerial image by an object detection model using machine learning

1. An apparatus for obtaining and geo-registering an aerial image of an object of interest, the apparatus comprising:

a memory configured to store computer-readable program code; and

processing circuitry configured to access the memory, and execute the computer-readable program code to cause the apparatus to at least:

obtain an aerial image by the processing circuitry being configured to receive the aerial image captured by a camera onboard an aircraft capturing the aerial image;

access an object detection model trained using a machine learning algorithm and a training set of aerial images of the object of interest having a defined boundary;

use the object detection model to detect the object of interest in the aerial image, the object detection including a prediction of a boundary of the object of interest depicted in the aerial image based on the defined boundary of the object of interest;

access a data store including a geographic location of the object of interest; and

geo-register the aerial image including the prediction of the boundary of the object of interest with the geographic location of the object of interest.

2. The apparatus of claim 1 , wherein the apparatus being caused to access the object detection model includes being caused to access the object detection model trained using a deep neural network of a deep learning algorithm.

3. The apparatus of claim 1 , wherein the apparatus being caused to use the object detection model to detect the object of interest includes being caused to use the object detection model to detect the object of interest including a pixelwise dense prediction of pixels of the object of interest depicted in the aerial image.

4. The apparatus of claim 1 , wherein the processing circuitry is configured to execute the computer-readable program code to cause the apparatus to further at least:

obtain the geographic location of the object of interest from a ground survey and independent of the training set of aerial images; and

create the data store with the geographic location so obtained.

5. The apparatus of claim 1 , wherein the apparatus being caused to geo-register the aerial image includes being caused to geo-register the aerial image through means of an embedded metadata tag that gives the geographic location of the object of interest.

6. The apparatus of claim 1 , wherein the camera or aircraft is equipped with a satellite-based navigation receiver configured to determine a geographic location of the object of interest,

wherein the apparatus being caused to obtain the aerial image includes being caused to obtain the aerial image with an embedded metadata tag that gives the geographic location of the object of interest determined by the satellite-based navigation receiver, and

wherein the apparatus being caused to geo-register the aerial image includes being caused to replace the embedded metadata tag that gives the geographic location of the object of interest determined by the satellite-based navigation receiver, with an embedded metadata tag that gives the geographic location of the object of interest from the data store.

7. The apparatus of claim 1 , wherein the training set on which the object detection model is trained further includes aerial images of a second object of interest having a second defined boundary,

wherein the apparatus being caused to use the object detection model includes being caused to use the object detection model to further detect the second object of interest in the aerial image, the object detection further including a pixelwise dense prediction of pixels of the aerial image in which the second object of interest is depicted based on the second defined boundary of the second object of interest,

wherein the apparatus being caused to access the data store includes being caused to access the data store further including a second geographic location of the second object of interest, and

wherein the apparatus being caused to geo-register the aerial image includes being caused to geo-register the aerial image further with the second geographic location of the second object of interest.

8. The apparatus of claim 1 , wherein the processing circuitry is configured to execute the computer-readable program code to cause the apparatus to further at least:

perform an affine transformation on the aerial image after the object detection model is used to detect the object of interest including the prediction of the boundary of the object of interest in the aerial image; and thereafter,

add the aerial image with the prediction of the boundary of the object of interest to the training set of aerial images.

9. A method of obtaining and geo-registering an aerial image of an object of interest, the method comprising:

obtaining an aerial image by processing circuitry receiving the aerial image captured by a camera onboard an aircraft capturing the aerial image;

accessing, by the processing circuitry, an object detection model trained using a machine learning algorithm and a training set of aerial images of the object of interest having a defined boundary; and by the processing circuitry,

using the object detection model to detect the object of interest in the aerial image, the object detection including a prediction of a boundary of the object of interest depicted in the aerial image based on the defined boundary of the object of interest;

accessing a data store including a geographic location of the object of interest; and

geo-registering the aerial image including the prediction of the boundary of the object of interest with the geographic location of the object of interest.

10. The method of claim 9 , wherein accessing the object detection model includes accessing the object detection model trained using a deep neural network of a deep learning algorithm.

11. The method of claim 9 , wherein using the object detection model to detect the object of interest includes using the object detection model to detect the object of interest including a pixelwise dense prediction of pixels of the object of interest depicted in the aerial image.

12. The method of claim 9 further comprising:

obtaining the geographic location of the object of interest from a ground survey and independent of the training set of aerial images; and

creating the data store with the geographic location so obtained.

13. The method of claim 9 , wherein geo-registering the aerial image includes geo-registering the aerial image through means of an embedded metadata tag that gives the geographic location of the object of interest.

14. The method of claim 9 , wherein the camera or aircraft is equipped with a satellite-based navigation receiver configured to determine a geographic location of the object of interest,

wherein obtaining the aerial image includes obtaining the aerial image with an embedded metadata tag that gives the geographic location of the object of interest determined by the satellite-based navigation receiver, and

wherein geo-registering the aerial image includes replacing the embedded metadata tag that gives the geographic location of the object of interest determined by the satellite-based navigation receiver, with an embedded metadata tag that gives the geographic location of the object of interest from the data store.

15. The method of claim 9 , wherein the training set on which the object detection model is trained further includes aerial images of a second object of interest having a second defined boundary,

wherein using the object detection model includes using the object detection model to further detect the second object of interest in the aerial image, the object detection further including a pixelwise dense prediction of pixels of the aerial image in which the second object of interest is depicted based on the second defined boundary of the second object of interest,

wherein accessing the data store includes accessing the data store further including a second geographic location of the second object of interest, and

wherein geo-registering the aerial image includes geo-registering the aerial image further with the second geographic location of the second object of interest.

16. The method of claim 9 further comprising:

performing an affine transformation on the aerial image after using the object detection model to detect the object of interest including the prediction of the boundary of the object of interest in the aerial image; and thereafter,

adding the aerial image with the prediction of the boundary of the object of interest to the training set of aerial images.

17. A non-transitory computer-readable storage medium for obtaining and geo-registering an aerial image of an object of interest, and having computer-readable program code stored therein that in response to execution by processing circuitry, causes an apparatus to at least:

obtain an aerial image by the processing circuitry being configured to receive the aerial image captured by a camera onboard an aircraft capturing the aerial image;

access an object detection model trained using a machine learning algorithm and a training set of aerial images of the object of interest having a defined boundary;

use the object detection model to detect the object of interest in the aerial image, the object detection including a prediction of a boundary of the object of interest depicted in the aerial image based on the defined boundary of the object of interest;

access a data store including a geographic location of the object of interest; and

geo-register the aerial image including the prediction of the boundary of the object of interest with the geographic location of the object of interest.

18. The computer-readable storage medium of claim 17 , wherein the apparatus being caused to access the object detection model includes being caused to access the object detection model trained using a deep neural network of a deep learning algorithm.

19. The computer-readable storage medium of claim 17 , wherein the apparatus being caused to use the object detection model to detect the object of interest includes being caused to use the object detection model to detect the object of interest including a pixelwise dense prediction of pixels of the object of interest depicted in the aerial image.

20. The computer-readable storage medium of claim 17 having further computer-readable program code stored therein that in response to execution by the processing circuitry causes the apparatus to further at least:

obtain the geographic location of the object of interest from a ground survey and independent of the training set of aerial images; and

create the data store with the geographic location so obtained.

21. The computer-readable storage medium of claim 17 , wherein the apparatus being caused to geo-register the aerial image includes being caused to geo-register the aerial image through means of an embedded metadata tag that gives the geographic location of the object of interest.

22. The computer-readable storage medium of claim 17 , wherein the camera or aircraft is equipped with a satellite-based navigation receiver configured to determine a geographic location of the object of interest,

wherein the apparatus being caused to obtain the aerial image includes being caused to obtain the aerial image with an embedded metadata tag that gives the geographic location of the object of interest determined by the satellite-based navigation receiver, and

wherein the apparatus being caused to geo-register the aerial image includes being caused to replace the embedded metadata tag that gives the geographic location of the object of interest determined by the satellite-based navigation receiver, with an embedded metadata tag that gives the geographic location of the object of interest from the data store.

23. The computer-readable storage medium of claim 17 , wherein the training set on which the object detection model is trained further includes aerial images of a second object of interest having a second defined boundary,

wherein the apparatus being caused to use the object detection model includes being caused to use the object detection model to further detect the second object of interest in the aerial image, the object detection further including a pixelwise dense prediction of pixels of the aerial image in which the second object of interest is depicted based on the second defined boundary of the second object of interest,

wherein the apparatus being caused to access the data store includes being caused to access the data store further including a second geographic location of the second object of interest, and

wherein the apparatus being caused to geo-register the aerial image includes being caused to geo-register the aerial image further with the second geographic location of the second object of interest.

24. The computer-readable storage medium of claim 17 having further computer-readable program code stored therein that in response to execution by the processing circuitry causes the apparatus to further at least:

perform an affine transformation on the aerial image after the object detection model is used to detect the object of interest including the prediction of the boundary of the object of interest in the aerial image; and thereafter,

add the aerial image with the prediction of the boundary of the object of interest to the training set of aerial images.