Automatic perspective transformation

1. A method comprising:

obtaining an image of a scene from a first perspective, the image including an object;

detecting the object in the image using a machine learning process, the object including a known shape with at least four vertices at a first set of points;

automatically predicting a second set of points corresponding to the at least four vertices of the object in a second perspective of the scene based on the known shape of the object and a pixel relative distance between the first set of points based on the known shape; and

constructing, without user input, a transformation matrix to transform a given image from the first perspective to the second perspective based on the first set of points and the second set of points.

2. The method of claim 1 , further comprising:

detecting a second object in the image using the machine learning process, the second object including a second known shape with at least four second object vertices at a third set of points;

automatically predicting a fourth set of points corresponding to the at least four second object vertices in the second perspective of the scene based on the second known shape of the second object;

constructing, without the user input, a second transformation matrix to transform the given image from the first perspective to the second perspective based on the third set of points and the fourth set of points; and

averaging the transformation matrix and the second transformation matrix to obtain a final transformation matrix of the image.

3. The method of claim 2 , further comprising:

obtaining a second image of the scene from the first perspective, the second image including the object and the second object;

detecting the object and the second object in the second image using the machine learning process, including identifying a fifth set of points corresponding to the at least four vertices of the object in the second image and a seventh set of points corresponding to the at least four second object vertices of the second object in the second image;

automatically predicting a sixth set of points corresponding to the at least four vertices of the object in the second perspective based on the known shape of the object;

automatically predicting an eighth set of points corresponding to the at least four second object vertices in the second perspective based on the second known shape of the second object;

constructing, without the user input, a third transformation matrix to transform the given image from the first perspective to the second perspective based on the fifth set of points and the sixth set of points;

constructing, without the user input, a fourth transformation matrix to transform the given image from the first perspective to the second perspective based on the seventh set of points and the eighth set of points;

averaging the third transformation matrix and the fourth transformation matrix to obtain a final transformation matrix of the second image; and

averaging the final transformation matrix of the image and the final transformation matrix of the second image to obtain a combined transformation matrix.

4. The method of claim 1 , wherein the image is obtained via a machine vision system, the method further comprising detecting a true distance between articles captured in the image based on the second perspective.

5. The method of claim 1 , wherein the machine learning process includes one of Mask region convolutional neural network (R-CNN) or You Only Look At CoefficienTs (YOLACT) instance segmentation to identify each instance of articles in the image.

6. The method of claim 1 , wherein obtaining the image includes capturing one or more images using a monocular camera.

7. The method of claim 1 , wherein the second perspective includes one of a top-down perspective or a front-view perspective.

8. One or more non-transitory computer readable media containing instructions that, when executed by one or more processors, cause a system to perform operations, the operations comprising:

obtaining an image of a scene from a first perspective, the image including an object;

detecting the object in the image using a machine learning process, the object including a known shape with at least four vertices at a first set of points;

automatically predicting a second set of points corresponding to the at least four vertices of the object in a second perspective of the scene based on the known shape of the object and a pixel relative distance between the first set of points based on the known shape; and

constructing, without user input, a transformation matrix to transform a given image from the first perspective to the second perspective based on the first set of points and the second set of points.

9. The one or more non-transitory computer readable media of claim 8 , the operations further comprising:

detecting a second object in the image using the machine learning process, the second object including a second known shape with at least four second object vertices at a third set of points;

automatically predicting a fourth set of points corresponding to the at least four second object vertices in the second perspective of the scene based on the second known shape of the second object;

constructing, without the user input, a second transformation matrix to transform the given image from the first perspective to the second perspective based on the third set of points and the fourth set of points; and

averaging the transformation matrix and the second transformation matrix to obtain a final transformation matrix of the image.

10. The one or more non-transitory computer readable media of claim 9 , the operations further comprising:

obtaining a second image of the scene from the first perspective, the second image including the object and the second object;

detecting the object and the second object in the second image using the machine learning process, including identifying a fifth set of points corresponding to the at least four vertices of the object in the second image and a seventh set of points corresponding to the at least four second object vertices of the second object in the second image;

automatically predicting a sixth set of points corresponding to the at least four vertices of the object in the second perspective based on the known shape of the object;

automatically predicting an eighth set of points corresponding to the at least four second object vertices in the second perspective based on the second known shape of the second object;

constructing, without the user input, a third transformation matrix to transform the given image from the first perspective to the second perspective based on the fifth set of points and the sixth set of points;

constructing, without the user input, a fourth transformation matrix to transform the given image from the first perspective to the second perspective based on the seventh set of points and the eighth set of points;

averaging the third transformation matrix and the fourth transformation matrix to obtain a final transformation matrix of the second image; and

averaging the final transformation matrix of the image and the final transformation matrix of the second image to obtain a combined transformation matrix.

11. The one or more non-transitory computer readable media of claim 8 , wherein the image is obtained via a machine vision system, the operations further comprising detecting a true distance between articles captured in the image based on the second perspective.

12. The one or more non-transitory computer readable media of claim 8 , wherein the machine learning process includes one of Mask region convolutional neural network (R-CNN) or You Only Look At CoefficienTs (YOLACT) instance segmentation to identify each instance of articles in the image.

13. The one or more non-transitory computer readable media of claim 8 , wherein obtaining the image includes capturing one or more images using a monocular camera.

14. The one or more non-transitory computer readable media of claim 8 , wherein the second perspective includes one of a top-down perspective or a front-view perspective.

15. A machine vision system comprising:

a monocular camera;

one or more processors; and

one or more non-transitory computer readable media containing instructions that, when executed by the one or more processors, cause the machine vision system to perform operations, the operations comprising:

obtaining an image of a scene from a first perspective as captured by the monocular camera, the image including an object;

detecting the object in the image using a machine learning process, the object including a known shape with at least four vertices at a first set of points;

automatically predicting a second set of points corresponding to the at least four vertices of the object in a second perspective of the scene based on the known shape of the object and a pixel relative distance between the first set of points based on the known shape;

constructing, without user input, a transformation matrix to transform a given image from the first perspective to the second perspective based on the first set of points and the second set of points; and

outputting a transformed image from the second perspective based on applying the transformation matrix to pixels of the image.

16. The machine vision system of claim 15 , the operations further comprising:

detecting a second object in the image using the machine learning process, the second object including a second known shape with at least four second object vertices at a third set of points;

automatically predicting a fourth set of points corresponding to the at least four second object vertices in the second perspective of the scene based on the second known shape of the second object;

constructing, without the user input, a second transformation matrix to transform the given image from the first perspective to the second perspective based on the third set of points and the fourth set of points; and

averaging the transformation matrix and the second transformation matrix to obtain a final transformation matrix of the image.

17. The machine vision system of claim 16 , the operations further comprising:

obtaining a second image of the scene from the first perspective from the monocular camera, the second image including the object and the second object;

detecting the object and the second object in the second image using the machine learning process, including identifying a fifth set of points corresponding to the at least four vertices of the object in the second image and a seventh set of points corresponding to the at least four second object vertices of the second object in the second image;

automatically predicting a sixth set of points corresponding to the at least four vertices of the object in the second perspective based on the known shape of the object;

automatically predicting an eighth set of points corresponding to the at least four second object vertices in the second perspective based on the second known shape of the second object;

constructing, without the user input, a third transformation matrix to transform the given image from the first perspective to the second perspective based on the fifth set of points and the sixth set of points;

constructing, without the user input, a fourth transformation matrix to transform the given image from the first perspective to the second perspective based on the seventh set of points and the eighth set of points;

averaging the third transformation matrix and the fourth transformation matrix to obtain a final transformation matrix of the second image; and

averaging the final transformation matrix of the image and the final transformation matrix of the second image to obtain a combined transformation matrix.