Systems and methods for recognizing text of interest

Patent metadata

Field	Value
Publication number	US-12374137-B2
Application number	US-202418616005-A
Country	US
Kind code	B2
Filing date	Mar 25, 2024
Priority date	Jul 20, 2021
Publication date	Jul 29, 2025
Grant date	Jul 29, 2025

Abstract

In some embodiments, apparatuses and methods are provided herein useful to determine text on an object. In some embodiments, there is provided a system to determine text of interest on an object of interest including a control circuit configured to execute a machine learning model trained to identify the text of interest, group into a cluster each character in the text of interest located substantially in the same location in the text of interest, determine a score value of each particular character in the cluster, identify the particular character that has a determined score value corresponding to at least a threshold score value relative to all characters in the cluster, assign the particular character having the determined score value corresponding to at least the threshold score value as a recognized character in the cluster, and output data comprising each recognized character associated with each cluster in the text of interest.

First claim

What is claimed is:

1. A pattern agnostic optical character recognition (OCR) system for determining text of interest on an object of interest, the pattern agnostic OCR system comprising:
a control circuit configured to execute a machine learning model trained to:
identify the text of interest on the object of interest in each captured digital image of a plurality of captured digital images of a scene;
group into a cluster each character in the text of interest located substantially in the same location in the text of interest in each captured digital image of the plurality of captured digital images of the scene;
determine a score value of each particular character in the cluster based on at least one of sum of a number of occurrences of the particular character in the cluster relative to a total sum of occurrences of all characters in the cluster and an average confidence value associated with the particular character;
identify the particular character that has a determined score value corresponding to at least a threshold score value relative to all characters in the cluster;
assign the particular character having the determined score value corresponding to at least the threshold score value as a recognized character in the cluster; and
output data comprising each recognized character associated with each cluster in the text of interest.

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

3. The pattern agnostic OCR system of claim 1 , wherein the text of interest comprises a corresponding identification associated with the object.

4. The pattern agnostic OCR system of claim 1 , wherein the character comprises one of a letter, a symbol of an alphabet, or a number, wherein the sum of the number of occurrences of the particular character in the cluster is weighted, and wherein the captured digital image was captured by at least one camera.

5. The pattern agnostic OCR system of claim 1 , wherein each character in the text of interest is vertically aligned relative to other characters in the text of interest.

6. The pattern agnostic OCR system of claim 5 , wherein the text of interest comprises one or more characters to be identified, wherein each character in the text of interest corresponds to a node point, and wherein the cluster comprises at least five node points to be grouped into the cluster.

7. The pattern agnostic OCR system of claim 1 , wherein the text of interest comprises one or more characters to be identified, and wherein each character in the text of interest corresponds to a node point, and wherein each character in the text of interest is horizontally aligned relative to other characters in the text of interest.

8. The pattern agnostic OCR system of claim 7 , wherein the cluster comprises at least four node points to be grouped into the cluster.

9. The pattern agnostic OCR system of claim 1 , wherein the text of interest comprises one or more characters to be identified, and wherein each character in the text of interest corresponds to a node point, wherein the control circuit is further configured to determine that a plurality of node points are located substantially in the same location in the text of interest of the captured digital image, and wherein the plurality of node points are grouped into the cluster when each node point of the plurality of node points is not more than seven unit distance from other node points in the plurality of node points.

10. The pattern agnostic OCR system of claim 9 , wherein the control circuit is further configured to:
determine that a plurality of clusters are located substantially in the same location in the text of interest of the captured digital image; and
merge two or more clusters of the plurality of clusters based on a unit distance between each cluster of the two or more clusters being less than a threshold merging value.

11. The pattern agnostic OCR system of claim 10 , wherein the threshold merging value is based on a ratio between a predetermined constant value associated with a total number of clusters associated with the text of interest and an average cluster distance between each cluster of the two or more clusters.

12. The pattern agnostic OCR system of claim 1 , further comprising a database configured to store a plurality of threshold merging values, each threshold merging value is associated with a possible total number of clusters in the text of interest and usable when a plurality of node points are located substantially in the same location in the text of interest.

13. The pattern agnostic OCR system of claim 1 , wherein the threshold score value corresponds to a highest score value relative to determined score values of all characters in the cluster.

14. A method for pattern agnostic optical character recognition (OCR) for determining text of interest on an object of interest, the method comprising:
identifying, by a control circuit, text of interest on the object of interest in each captured digital image of a plurality of captured digital images of a scene;
grouping, by the control circuit, into a cluster each character in the text of interest located substantially in the same location in the text of interest in each captured digital image of the plurality of captured digital images of the scene;
determining, by the control circuit, a score value of each particular character in the cluster based on a sum of a number of occurrences of the particular character in the cluster relative to a total sum of occurrences of all characters in the cluster and an average confidence value associated with the particular character;
identifying, by the control circuit, the particular character that has a determined score value corresponding to at least a threshold score value relative to all characters in the cluster;
assigning, by the control circuit, the particular character having the determined score value corresponding to at least the threshold score value as a recognized character in the cluster;
outputting, by the control circuit, each recognized character associated with each cluster in the text of interest.

15. The method of claim 14 , wherein the object comprises a cargo trailer for use in distribution of goods, the text of interest identifying the cargo trailer.

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

17. The method of claim 14 , wherein the character comprises a letter, a symbol of an alphabet, and a number, wherein the sum of the number of occurrences of the particular character in the cluster is weighted, and wherein the captured digital image was captured by at least one camera.

18. The method of claim 14 , wherein each character in the text of interest is vertically aligned relative to other characters in the text of interest.

19. The method of claim 18 , wherein the text of interest comprises one or more characters to be identified, wherein each character in the text of interest corresponds to a node point, and wherein the cluster comprises at least five node points to be grouped into the cluster.

20. The method of claim 14 , wherein the text of interest comprises one or more characters to be identified, and wherein each character in the text of interest corresponds to a node point, and wherein each character in the text of interest is horizontally aligned relative to other characters in the text of interest.

Assignees

• Walmart Apollo Llc

Inventors

• RAMASWAMY SRINIVASA Ramanujam
• Kumar Manish
• AGGARWAL Pranav

Classifications

• G06F18/23

  Clustering techniques · CPC title

• G06V20/62

  Text, e.g. of license plates, overlay texts or captions on TV images · CPC title

• G06T11/60

  Creating or editing images; Combining images with text · CPC title

• G06V10/82

  using neural networks · CPC title

• G06V2201/08

  Detecting or categorising vehicles · CPC title