# US 2020/0401833 A1 · License plate detection system

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2020/0401833 A1
- Document ID: US20200401833A1
- Status: Published application
- Kind: A1
- Application number: 16/444,941
- Filed: 2019-06-18
- Published: 2020-12-24
- Earliest claimed: 2019-06-18
- Assignees: Nortek Security & Control LLC
- Applicants: Nortek Security & Control LLC.
- Inventors: Ilya Popov; Krishna Khadloya; Sofiya Klyan
- Web page: https://patentgrep.com/patent/US20200401833A1

## CPC

- G06V 10/764: using classification, e.g. of video objects
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques}
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/09: Supervised learning
- G06V 10/7784: {based on feedback from supervisors}
- G06V 10/82: using neural networks
- G06V 20/41: {Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items (segmenting video sequences G06V20/49)}
- G06V 20/63: {Scene text, e.g. street names}
- G06N 3/045: Combinations of networks
- G06V 20/625: {License plates}
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A system for detecting license plates is described. The system receives raw data comprising images of license plates. A base version of a ground truth is prepared based on the raw data, using a generic license plate detection (LPD). The system prepares input data for training a deep learning network. The deep learning network is trained with the prepared input data. A newly trained generic LPD is formed using data generated by the existing generic LPD.

## Claims

### Claim 1

A computer-implemented method comprising:
receiving raw data comprising images of license plates;
preparing, using a generic license plate detection (LPD), a base version of a ground truth based on the raw data;
preparing input data for training a deep learning network;
training the deep learning network with the prepared input data; and
forming a newly trained generic LPD with the trained deep learning network.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, further comprising:
testing the newly trained generic LPD.

### Claim 3 (depends on 1)

The computer-implemented method of claim 1, wherein the raw data further comprises license plates from a plurality of countries, bounding boxes, and vehicle types.

### Claim 4 (depends on 1)

The computer-implemented method of claim 1, wherein preparing the base version of the ground truth further comprises:
operating the generic LPD on the raw data to generate bounding boxes, classes, and country tags, the raw data further comprising existing information about bounding boxes, classes, and country tags; and
validating, using the generic LPD, the generated bounding boxes, classes, and country tags with the existing information from the raw data.

### Claim 5 (depends on 1)

The computer-implemented method of claim 1, wherein preparing the input data further comprises:
gathering media data for a plurality of countries;
forming an image of a license plate from a video frame;
collecting the gathered media data separately for single row license plates and multiline license plates;
splitting the single row and the multiline sets of data into training and validation sets; and
saving the training and validation sets in a specific format.

### Claim 6 (depends on 1)

The computer-implemented method of claim 1, wherein the deep learning network receives an input image and outputs a bounding box corresponding to a license plate in the input image.

### Claim 7 (depends on 1)

The computer-implemented method of claim 1, wherein an output for the ground truth for license plate further comprises: initially autogenerated GT for license plates at images, and validation and suggestions for modifying of already existing GT for license plates.

### Claim 8 (depends on 1)

The computer-implemented method of claim 1, further comprising:
updating the generic LPD with the newly trained LPD.

### Claim 9 (depends on 1)

The computer-implemented method of claim 1, further comprising:
forming a customized LPD for each country.

### Claim 10 (depends on 1)

The computer-implemented method of claim 1, wherein the deep neural network comprises a convolutional neural network (CNN).

### Claim 11

A computing apparatus, the computing apparatus comprising:
a processor; and
a memory storing instructions that, when executed by the processor, configure the apparatus to:
receive raw data comprising images of license plates;
prepare, using a generic license plate detection (LPD), a base version of a ground truth based on the raw data;
prepare input data for training a deep learning network;
train the deep learning network with the prepared input data; and
form a newly trained generic LPD with the trained deep learning network.

### Claim 12 (depends on 11)

The computing apparatus of claim 11, wherein the instructions further configure the apparatus to:
test the newly trained generic LPD.

### Claim 13 (depends on 11)

The computing apparatus of claim 11, wherein the raw data further comprises license plates from a plurality of countries, bind boxes, and vehicle types.

### Claim 14 (depends on 11)

The computing apparatus of claim 11, wherein preparing the base version of the ground truth further comprises:
operate the generic LPD on the raw data to generate bounding boxes, classes, and country tags, the raw data further comprising existing information about bounding boxes, classes, and country tags; and
validate, using the generic LPD, the generated bounding boxes, classes, and country tags with the existing information from the raw data.

### Claim 15 (depends on 11)

The computing apparatus of claim 11, wherein preparing the input data further comprises:
gathering media data for a plurality of countries;
form an image of a license plate from a video frame;
collect the gathered media data separately for single row license plates and multiline license plates;
split the single row and the multiline sets of data into training and validation sets; and
save the training and validation sets in a specific format.

### Claim 16 (depends on 11)

The computing apparatus of claim 11, wherein the deep learning network receives an input image and outputs a bounding box corresponding to a license plate in the input image.

### Claim 17 (depends on 11)

The computing apparatus of claim 11, wherein an output for the ground truth for license plate further comprises: initially autogenerated GT for license plates at images, and validation and suggestions for modifying of already exist GT for license plates.

### Claim 18 (depends on 11)

The computing apparatus of claim 11, wherein the instructions further configure the apparatus to:
update the generic LPD with the newly trained LPD.

### Claim 19 (depends on 11)

The computing apparatus of claim 11, wherein the instructions further configure the apparatus to:
form a customized LPD for each country.

### Claim 20

A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computer, cause the computer to:
receive raw data comprising images of license plates;
prepare, using a generic license plate detection (LPD), a base version of a ground truth based on the raw data;
prepare input data for training a deep learning network;
train the deep learning network with the prepared input data; and
form a newly trained generic LPD with the trained deep learning network.