Automatic learning for vehicle classification

Patent metadata

Field	Value
Publication number	US-10657809-B2
Application number	US-201615772106-A
Country	US
Kind code	B2
Filing date	Dec 29, 2016
Priority date	Dec 30, 2015
Publication date	May 19, 2020
Grant date	May 19, 2020

Abstract

The present disclosure provides an automatically updating vehicle classification system and method. The system comprises a processor which extracts from a vehicle image at least one of: a unique vehicle identifier from the vehicle image and visual features of the vehicle in the vehicle image. If the visual features are below a probability threshold for matching a vehicle class in a local database, the processor looks up the unique vehicle identifier in a registration database. The registration database stores vehicle registration information including unique vehicle identifiers and associated vehicle class information. If the vehicle class information associated with the vehicle identifier is not a class recognized by the processor, the processor creates a new vehicle class associated with the visual features of the vehicle.

First claim

What is claimed is:

1. An automatically updating vehicle classification system, comprising:
a processor which extracts from a vehicle image a unique vehicle identifier from the vehicle image and visual features of the vehicle in the vehicle image;
wherein if the visual features are below a probability threshold for matching a vehicle class in a local database, the processor looks up the unique vehicle identifier in a registration database;
wherein the registration database stores vehicle registration information including unique vehicle identifiers and associated vehicle class information;
wherein if the vehicle class information associated with the vehicle identifier is not a class recognized by the processor, the processor creates a new vehicle class associated with the visual features of the vehicle; and
wherein the probability threshold is determined based on sample data sets to determine where matching accuracy is below a particular level.

2. The system of claim 1 , wherein vehicle class information includes at least one of or any combination of: make, model, body type, seat capacity, engine size, manufacture date, number of axles, and color.

3. The system of claim 1 , wherein the new vehicle class is not activated for use until the processor has looked up a predetermined minimum number of unique vehicle identifiers associated with the new vehicle class.

4. The system of claim 1 , further comprising a camera that captures the vehicle image.

5. The system of claim 1 , wherein the processor looks up the unique vehicle identifier in the registration database during a low volume time period.

6. The system of claim 1 , wherein if the new class has only a single vehicle in it for a defined period of time, the single vehicle is designated as an outlier.

7. The system of claim 6 , wherein the processor designates outliers for manual review.

8. The system of claim 1 , wherein the processor creates new classes while the system is deployed for use.

9. The system of claim 1 , wherein the processor updates classes while the system is deployed for use.

10. The system of claim 1 , wherein the unique vehicle identifier is displayed on a number plate.

11. A method for automatically updating a vehicle classification system, comprising:
providing a vehicle image;
extracting, with a processor, a unique vehicle identifier from the vehicle image and visual features of the vehicle in the vehicle image; and
if the visual features of the vehicle are below a probability threshold for matching a vehicle class in a local database, looking up, with the processor, the unique vehicle identifier in a registration database;
wherein the registration database stores registration information including unique vehicle identifiers and associated vehicle class information;
wherein if the vehicle class information associated with the unique vehicle identifier is not a vehicle class recognized by the processor, creating, with the processor a new vehicle class associated with the visual features of the vehicle; and
wherein the probability threshold is determined based on sample data sets to determine where matching accuracy is below a particular level.

12. The method of claim 11 , wherein vehicle class information includes at least one of or any combination of: make, model, body type, number of axles, seat capacity, engine size, manufacture date and color.

13. The method of claim 11 , wherein the new vehicle class is not activated for use until the processor has looked up a minimum number of unique vehicle identifiers associated with the new class.

14. The method of claim 11 , further comprising capturing, with a camera, the vehicle image.

15. The method of claim 11 , wherein the processor looks up the unique vehicle identifier in the registration database during a low volume time period.

16. The method of claim 11 , further comprising designating the vehicle as an outlier if the new class has only the single vehicle in it for a defined period of time.

17. The method of claim 16 , further comprising designating the outliers for manual review.

18. The method of claim 11 , further comprising creating, with the processor, new vehicle classes while the system is deployed for use.

19. The method of claim 11 , wherein the processor updates vehicle classes while the system is deployed for use.

20. The method of claim 11 , wherein the unique vehicle identifier is displayed on a number plate.

Assignees

• 3M Innovative Properties Co

Inventors

• Schumacher Stephanie R
• Smith Brian D

Classifications

• G06N20/00 Primary

  Machine learning · CPC title

• G06K9/00785

  Physics · mapped topic

• G06K9/325

  Physics · mapped topic

• G08G1/0175 Primary

  by photographing vehicles, e.g. when violating traffic rules · CPC title

• G06K9/6263

  Physics · mapped topic