Methods and systems for enhancing the performance of automated license plate recognition applications utilizing multiple results

• US 8,781,172 B2
• Filed: 03/30/2012
• Issued: 07/15/2014
• Est. Priority Date: 03/30/2012
• Status: Active Grant

• Current Assignee: Conduent Business Services, LLC (Conduent, Inc.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Kozitsky, Vladimir, Burry, Aaron Michael
• Primary Examiner(s): Tabatabai, Abolfazl
• Application Number:
  US13/435,875
• Publication Number:
  US 20130259314A1
• Time in Patent Office: 837 Days
• Field of Search: 382/103, 382/104, 382/105, 382/236, 348/148, 348/149, 348/169, 348/170, 348/171, 348/172, 340/2.1, 340/5.1, 340/10.1, 340/928, 340/933, 340/937, 340/988, 362/497
• US Class Current: 382/105
• CPC Class Codes: G06V 20/625 License plates
• CPC Class Codes: G06V 20/63 Scene text, e.g. street names

Abstract

Methods, systems and processor-readable media for enhancing performance of an automated license plate recognition system utilizing multiple results. Multiple images can be captured as a vehicle passes through an observation zone and each image can be processed utilizing an ALPR unit to obtain a plate code result and associated confidence values. Iterative processing of character level information across an OCR code followed by a higher level error checking based on learned context information can be performed. A string correlation approach can be employed to optimally align the OCR code from multiple images despite noise factors. The OCR confidence and state mask information can then be leveraged to select a character for an output plate code taking into account the ALPR error sources.

Claims

1. A method for enhancing the performance of an automated license plate recognition unit, said method comprising:

capturing a plurality of images with respect to at least one vehicle passing through an observation zone and processing each image among said plurality of images utilizing an automatic license plate recognition unit to obtain a plate code result and a confidence value;

iteratively processing character level information across an optical character recognition code followed by a higher level error checking operation based on a learned context information; and

optimally aligning said optical character recognition code with respect to said plurality of images despite a noise factor utilizing a string correlation approach to thereafter leverage an optical character recognition confidence and state mask information with respect to said plate code result and said confidence value to select a character for said output plate code while taking into account an error source so as to yield more accurate results thereof.

2. The method of claim 1 wherein said plurality of images comprises images of at least one of:

a front license plate image; and

a rear license plate image.

3. The method of claim 1 wherein iteratively processing said character level information, further comprises:

parsing through an optical character recognition result and identifying a common pattern between said result utilizing said string correlation approach;

recording a plurality of matching characters between said codes at each index value by sliding result strings past one another location by location; and

identifying an index value which provides a most common element between said strings utilizing a correlation information and thereafter aligning said strings based on an optimal offset to partition said code into said common pattern and a residual block.

4. The method of claim 3 further comprising:

identifying a non-matching location in said common pattern and selecting a highest confidence optical character recognition result to resolve an ambiguity after partitioning said code string;

processing each remaining residual block utilizing said correlation analysis and setting an error flag if said residual block for both codes are of different lengths and have no characters in common; and

terminating said algorithm if all characters are labeled as complete.

5. The method of claim 1 further comprising:

applying a secondary error checking if said low-level character by character processing is complete to leverage information regarding a valid license plate sequence for a plurality of states; and

determining a jurisdiction of said at least one vehicle by utilizing a plate mask, a placement of character in a particular position, and a frequency that each mask and character combination appears in a training set as input to a machine learning module.

6. The method of claim 5 further comprising:

determining a likely state by said machine learning module and a likely plate code by a confidence associated with a conclusion;

sorting decisions made at a character level from a lowest margin to a highest margin above an optical character recognition threshold to undo a change starting with a most likely wrong; and

continuing said process as long as a state identification module returns a low confidence for an overall result.

7. The method of claim 1, further comprising determining said automated license plate recognition result utilizing at least three confidence values.

8. The method of claim 7 wherein said at least three confidence values comprises at least one of the following types of confidence values:

an optical character recognition confidence for an inside character and a character bounded by other valid character;

an optical character recognition for a border character and a first and a last character of said plate code; and

a state identification confidence.

9. A system for enhancing the performance of an automated license plate recognition unit, said system comprising:

a processor;

a data bus coupled to said processor; and

a computer-usable medium embodying computer program code, said computer-usable medium being coupled to said data bus, said computer program code comprising instructions executable by said processor and configured for;

capturing a plurality of images with respect to at least one vehicle passing through an observation zone and processing each image among said plurality of images utilizing an automatic license plate recognition unit to obtain a plate code result and a confidence value;

iteratively processing character level information across an optical character recognition code followed by a higher level error checking operation based on a learned context information; and

optimally aligning said optical character recognition code with respect to said plurality of images despite a noise factor utilizing a string correlation approach to thereafter leverage an optical character recognition confidence and state mask information with respect to said plate code result and said confidence value to select a character for said output plate code while taking into account an error source so as to yield more accurate results thereof.

10. The system of claim 9 wherein said plurality of images comprises of images of at least one of:

a front license plate image and a rear license plate image.

11. The system of claim 9 wherein said instructions for iteratively processing said character level information are further configured to comprise instructions for:

parsing through an optical character recognition result and identifying a common pattern between said result utilizing said string correlation approach;

recording a plurality of matching characters between said codes at each index value by sliding result strings past one another location by location; and

identifying an index value which provides a most common element between said strings utilizing a correlation information and thereafter aligning said strings based on an optimal offset to partition said code into said common pattern and a residual block.

12. The system of claim 11 wherein said instructions are further configured for:

identifying a non-matching location in said common pattern and selecting a highest confidence optical character recognition result to resolve an ambiguity after partitioning said code string;

processing each remaining residual block utilizing said correlation analysis and setting an error flag if said residual block for both codes are of different lengths and have no characters in common; and

terminating said algorithm if all characters are labeled as complete.

13. The system of claim 9 wherein said instructions are further configured for:

applying a secondary error checking if said low-level character by character processing is complete to leverage information regarding a valid license plate sequence for a plurality of states; and

determining a jurisdiction of said at least one vehicle by utilizing a plate mask, a placement of character in a particular position, and a frequency that each mask and character combination appears in a training set as input to a machine learning module.

14. The system of claim 13 wherein said instructions are further configured for:

determining a likely state by said machine learning module and a likely plate code by a confidence associated with a conclusion;

sorting decisions made at a character level from a lowest margin to a highest margin above an optical character recognition threshold to undo a change starting with a most likely wrong; and

continuing said process as long as a state identification module returns a low confidence for an overall result.

15. The system of claim 9 wherein said instructions are further configured for determining said automated license plate recognition result utilizing at least three confidence values.

16. The system of claim 15 wherein said at least three confidence values comprises at least one of the following types of confidence values:

an optical character recognition confidence for an inside character and a character bounded by other valid character;

an optical character recognition for a border character and a first and a last character of said plate code; and

a state identification confidence.

17. A non-transitory processor-readable medium storing code representing instructions to cause a process to enhance the performance of an automated license plate recognition unit, said code comprising code to:

capture a plurality of images with respect to at least one vehicle passing through an observation zone and processing each image among said plurality of images utilizing an automatic license plate recognition unit to obtain a plate code result and a confidence value;

iteratively process character level information across an optical character recognition code followed by a higher level error checking operation based on a learned context information; and

optimally align said optical character recognition code with respect to said plurality of images despite a noise factor utilizing a string correlation approach to thereafter leverage an optical character recognition confidence and state mask information with respect to said plate code result and said confidence value to select a character for said output plate code while taking into account an error source so as to yield more accurate results thereof.

18. The non-transitory processor-readable medium of claim 17 wherein said plurality of images comprises images of at least one of:

a front license plate image; and

a rear license plate image.

19. The non-transitory processor-readable medium of claim 17 wherein said code further comprises code to:

parse through an optical character recognition result and identify a common pattern between said result utilizing said string correlation approach;

record a plurality of matching characters between said codes at each index value by sliding result strings past one another location by location; and

identify an index value which provides a most common element between said strings utilizing a correlation information and thereafter aligning said strings based on an optimal offset to partition said code into said common pattern and a residual block.

20. The non-transitory processor-readable medium of claim 19 wherein said code further comprises code to:

identify a non-matching location in said common pattern and selecting a highest confidence optical character recognition result to resolve an ambiguity after partitioning said code string;

process each remaining residual block utilizing said correlation analysis and set an error flag if said residual block for both codes are of different lengths and have no characters in common; and

terminate said algorithm if all characters are labeled as complete.