Systems and methods for detecting text of interest

Patent metadata

Field	Value
Publication number	US-11948374-B2
Application number	US-202117380506-A
Country	US
Kind code	B2
Filing date	Jul 20, 2021
Priority date	Jul 20, 2021
Publication date	Apr 2, 2024
Grant date	Apr 2, 2024

Abstract

In some embodiments, apparatuses and methods are provided herein useful to train a machine learning algorithm to detect text of interest. In some embodiments, there is provided a system to detect vertically oriented text of interest including a first data set comprising a plurality of captured digital images each depicting an object of interest and a second data set comprising a plurality of augmented digital images each depicting a captured digital image augmented with a synthetic text image; a first control circuit configured to cause the machine learning algorithm to output a machine learning model trained to automatically detect occurrences of vertically oriented text of interest based on the first data set and the second data set; at least one camera; and a second control circuit configured to execute the machine learning model to automatically detect vertically oriented text of interest on the object of interest.

First claim

What is claimed is:

1. A system for training a machine learning algorithm to detect text of interest, the system comprising:
a first data set comprising a plurality of captured digital images each depicting an object of interest;
a second data set comprising a plurality of augmented digital images each depicting a captured digital image augmented with a synthetic text image comprising randomly generated text on a randomly selected background image;
a first control circuit configured to:
execute a machine learning algorithm stored in a memory;
cause the machine learning algorithm to be executed with the first data set as a first input;
cause the machine learning algorithm to be executed with the second data set as a second input; and
cause the machine learning algorithm to output a machine learning model trained to automatically detect occurrences of vertically oriented text of interest depicted in captured digital images based on the first data set and the second data set;

at least one camera configured to capture one or more digital images of a scene; and
a second control circuit coupled to the at least one camera and configured to execute the machine learning model to automatically detect vertically oriented text of interest on the object of interest depicted in the captured one or more digital images of the scene.

2. The system of claim 1 , wherein the object of interest comprises a cargo trailer for use in distribution of goods.

3. The system of claim 1 , wherein the vertically oriented text of interest comprises a corresponding identification associated with the object of interest.

4. The system of claim 1 , wherein the vertically oriented text of interest comprises text having a combination of one or more alphabet and numerical characters.

5. The system of claim 1 , wherein the synthetic text image comprises text that is randomly oriented.

6. The system of claim 1 , wherein the second control circuit is further configured to:
determine a first image resolution of the captured one or more digital images;
determine that the first image resolution is less than a threshold image resolution; and
in response to the determination that the first image resolution is less than the threshold image resolution, pad the captured one or more digital images with images to match the threshold image resolution, wherein the padding of the captured one or more digital images with the images avoids stretching the captured one or more digital images to match the threshold image resolution and facilitates a first differentiation between letter O and number 0 and a second differentiation between letter I and number 1.

7. The system of claim 1 , wherein the first control circuit is further configured to determine the synthetic text image based on:
a first selection of a background image;
a second selection of an area in the background image;
a generation of the text; and
a third selection of an orientation of the text.

8. The system of claim 7 , wherein one or more of the first selection, the second selection, the generation and the third selection are randomly determined by the first control circuit.

9. The system of claim 7 , wherein the first control circuit is further configured to determine a font color of the generated text based on:
a determination of red, green, and blue (RGB) channel distributions of the selected area in the background image;
a determination of a set of least participating values in each RGB channel;
randomly select a value in each set associated with each RGB channel; and
determine the font color based on the randomly selected value in each set associated with each RGB channel.

10. The system of claim 9 , wherein the set of least participating values in each RGB channel comprises a range of values not represented in a corresponding distribution of each RGB channel.

11. The system of claim 1 , further comprising a third data set comprising a plurality of captured digital images each depicting the object of interest having vertically oriented text of interest, wherein the first control circuit is further configured to cause the machine learning algorithm to be executed with the third data set as a third input.

12. A method for training a machine learning algorithm to detect text of interest, the method comprising:
executing, by a first control circuit coupled to a memory, a machine learning algorithm stored in the memory;
causing, by the first control circuit, the machine learning algorithm to be executed with a first data set as a first input, wherein the first data set comprises a plurality of captured digital images each depicting an object of interest;
causing, by the first control circuit, the machine learning algorithm to be executed with a second data set as a second input, wherein the second data set comprises a plurality of augmented digital images each depicting a captured digital image augmented with a synthetic text image comprising randomly generated text on a randomly selected background image;
causing, by the first control circuit, the machine learning algorithm to output a machine learning model trained to automatically detect occurrences of vertically oriented text of interest depicted in captured digital images based on the first data set and the second data set; and
executing, by a second control circuit coupled to at least one camera configured to capture one or more digital images of a scene, the machine learning model to automatically detect the vertically oriented text of interest on the object of interest in the captured one or more digital images of the scene.

13. The method of claim 12 , wherein the object of interest comprises a cargo trailer for use in distribution of goods.

14. The method of claim 12 , wherein the vertically oriented text of interest comprises a corresponding identification associated with the object of interest.

15. The method of claim 12 , wherein the vertically oriented text of interest comprises text having a combination of one or more alphabet and numerical characters.

16. The method of claim 12 , wherein the synthetic text image comprises text that is randomly oriented.

17. The method of claim 12 , further comprising:
determining, by the second control circuit, a first image resolution of the captured one or more digital images;
determining, by the second control circuit, that the first image resolution is less than a threshold image resolution; and
in response to the determination that the first image resolution is less than the threshold image resolution, padding, by the second control circuit, the captured one or more digital images with images to match the threshold image resolution, wherein the padding of the captured one or more digital images with the images avoids stretching the captured one or more digital images to match the threshold image resolution and facilitates a first differentiation between letter O and number 0 and a second differentiation between letter I and number 1.

18. The method of claim 12 , further comprising, in a determination of the synthetic text image:
first selecting, by the first control circuit, a background image;
second selecting, by the first control circuit, an area in the background image;
generating, by the first control circuit, text; and
third selecting, by the first control circuit, an orientation of the text.

19. The method of claim 18 , wherein one or more of the first selection, the second selection, the generation and the third selection are randomly determined by the first control circuit.

Assignees

• Walmart Apollo Llc

Inventors

• RAMASWAMY SRINIVASA Ramanujam
• Kumar Manish
• AGGARWAL Pranav

Classifications

• G06V20/63 Primary

  Scene text, e.g. street names · CPC title

• G06N20/00

  Machine learning · CPC title

• G06V30/1444

  Selective acquisition, locating or processing of specific regions, e.g. highlighted text, fiducial marks or predetermined fields · CPC title

• G06V30/1463

  Orientation detection or correction, e.g. rotation of multiples of 90 degrees · CPC title

• G06V20/625

  License plates · CPC title