System and method for reading license plate from an image using an ensemble method of ALPR systems

Patent metadata

Field	Value
Publication number	US-12482277-B2
Application number	US-202217881195-A
Country	US
Kind code	B2
Filing date	Aug 4, 2022
Priority date	Aug 4, 2022
Publication date	Nov 25, 2025
Grant date	Nov 25, 2025

Abstract

Methods and systems of license plate recognition can involve subjecting an image captured by an image capturing device to image-processing by a group of different license plate recognition engines including a license plate recognition engine and a license plate reidentification engine, and using a decision tree to combine data from the license plate recognition engine and the license plate reidentification engine and generate a license plate identifier based on the data processed by the decision tree.

First claim

What is claimed is:

1 . A method of license plate recognition, comprising: subjecting an image captured by an image capturing device to image-processing by a plurality of different license plate recognition engines including a license plate recognition engine and a license plate reidentification engine, the image capturing device comprising an Automatic License Plate Recognition (ALPR) camera and the license plate reidentification engine comprising a template-based ALPR solution that encodes the image and text found in the image using a plurality of neural networks including at least two different neural networks; outputting distance values of the top-3 template-based ALPR solution results and a confidence value of an output of the license plate recognition engine to construct at least a 16-dimensional vector which becomes an input to a decision tree; and using the decision tree to combine data output from the license plate recognition engine and the license plate reidentification engine and generate a license plate identifier based on the data input to and processed by the decision tree, the data input to the decision tree including an input vector.

2 . The method of claim 1 further comprising constructing the input vector to the decision tree by:
collecting multiple candidate license plate numbers from the template-based ALPR solution;
generating a set of feature vectors based on distance values associated with the candidate license plate numbers;
including an additional candidate license plate number from an OCR output if it is not already present among the template-based ALPR solution outputs;
constructing a feature vector for the OCR output based on its confidence value; and
combining the feature vectors into a single input vector that represents both the template-based ALPR solution and the OCR output, wherein the input vector is used by the decision tree to generate the license plate identifier.

3 . The method of claim 2 wherein the decision tree generates an output value corresponding to a single integer within a predefined range, wherein the integer serves as an index to a list of license plate number outputs from the license plate recognition engine and the license plate reidentification engine, and wherein the index selects one of the top-ranked results from the template-based ALPR solution or the OCR output as the license plate identifier.

4 . The method of claim 2 further comprising determining a nearest license plate match by calculating a distance between an image encoding generated by the template-based ALPR solution and each corresponding text encoding, wherein the distance is used to select a most likely license plate identifier.

5 . The method of claim 4 further comprising capturing the image with the ALPR camera, wherein the ALPR camera comprises a high speed camera with an infrared filter or at least two cameras including a high resolution digital camera and an infrared camera.

6 . The method of claim 1 further comprising constructing the input vector to the decision tree by:
generating feature vectors based on distance values from the template-based ALPR solution and confidence values from an OCR output; and
combining the feature vectors into a single input vector for processing by the decision tree.

7 . The method of claim 6 wherein the at least two different neural networks comprises at least two different shallow neural networks.

8 . The method of claim 6 further comprising using a distance of the image encoding from each text encoding to determine a nearest license plate match, wherein the distance comprises the distance values.

9 . A method of license plate recognition, comprising: subjecting an image captured by an image capturing device to image-processing by a plurality of different license plate recognition engines including a license plate recognition engine and a license plate reidentification engine, wherein the license plate reidentification engine comprises a template-based ALPR solution, the image capturing device comprising an Automatic License Plate Recognition (ALPR) camera comprising a high speed camera with an infrared filter or at least two cameras including a high resolution digital camera and an infrared camera; using a decision tree to combine data output from the license plate recognition engine and the license plate reidentification engine and generate a license plate identifier based on the data input to and processed by the decision tree, the data input to the decision tree including at least an 16-dimensional input vector; and constructing an input vector for the decision tree by processing outputs from the template-based ALPR solution and the license plate recognition engine, wherein the input vector includes distance values of top-ranked template-based ALPR solution results and confidence values from the license plate recognition engine; and using the decision tree to generate a license plate identifier by selecting an index corresponding to one of the top-ranked template-based ALPR outputs or an OCR-derived output, wherein the index represents the most probable license plate match.

10 . The method of claim 9 wherein the template-based ALPR solution encodes the image and text found in the image using a plurality of neural networks.

11 . The method of claim 10 wherein the plurality of neural networks comprises at least two different shallow neural networks.

12 . The method of claim 10 further comprising using a distance of the image encoding from each text encoding to determine a nearest license plate match.

13 . An ensemble automatic license plate recognition (ALPR) system of license plate recognition, comprising: an image capturing device comprising an ALPR camera including a high speed camera with an infrared filter or at least two cameras including a high resolution digital camera and an infrared camera, wherein an image captured by the image capturing device is subject to image-processing by a plurality of different license plate recognition engines including a license plate recognition engine and a license plate reidentification engine, wherein the license plate reidentification engine comprises a template-based ALPR solution that encodes the image and text found in the image using a plurality of neural network; a processor operable to construct an input vector for a decision tree by processing outputs from the template-based ALPR solution and the license plate recognition engine, wherein the input vector includes distance values of top-ranked template-based ALPR solution results and confidence values from the license plate recognition engine; and the decision tree configured to generate a license plate identifier by selecting an index corresponding to one of the top-ranked template-based ALPR outputs or an OCR-derived output, wherein the index represents the most probable license plate match, wherein the decision tree combines data from the license plate recognition engine and the license plate reidentification engine and generates a license plate identifier based on the data processed by the decision tree, the data input to the decision tree including the input vector comprising at least a 16-dimensional vector.

Assignees

• Conduent Business Services Llc

Inventors

• Gudipati Neeraj
• Gayatri Vinuta Vishweshwar

Classifications

• G06T2207/20084

  Artificial neural networks [ANN] · CPC title

• G06T2207/20081

  Training; Learning · CPC title

• G06T2207/30248

  Vehicle exterior or interior · CPC title

• G06T7/0004

  Industrial image inspection · CPC title

• G06V20/62

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