Pitfalls of Optical Character Recognition OCR Scanning for the Mississippi Uniform Crash Report Appl

1
Pitfalls of Optical Character Recognition (OCR)
Scanning for the Mississippi Uniform Crash Report
Application2005 Traffic Records ForumDonna
SmithProgrammer AnalystMississippi Department
of Public Safety
2
Overview

• In 2003, Mississippi Department of Public Safety
  and Mississippi Department of Transportation
  recognized a need to modernize its crash data
  collection mechanisms
• Initially, the State of MS chose to pursue an
  Optical Character Recognition (OCR) solution for
  electronically scanning data off of the paper
  reports and storing information

3
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report

4
The Officer on the Scene Fills in the Mississippi
Uniform Crash Report (MUCR)

• All officers in MS have paper versions of the
  report.
• The form was redesigned to capture information
  efficiently and completely through the use of
  pre-defined fields wherever possible.
• Intent is to facilitate the officer making quick
  selections through check boxes and option buttons
  in a multiple-choice fashion.

5
State of Mississippi Uniform Crash Report

• Always Page 1.
• Includes First Harmful Event, Crash location,
  Roadway System, etc.
• Reporting Officer Badge Number and Name.
• Reviewing Officer Badge Number and Name.

6
Mississippi Uniform Crash ReportDiagram /
Narrative

• Top part is dedicated to the hand drawn diagram.
• Bottom part is dedicated to the hand written
  narrative.

7
Mississippi Uniform Crash ReportPerson / Occupant

• Includes information about the driver.
• Includes information up to 2 occupants.
• Items included are Driver Condition, Contributing
  Circumstances, Safety Equipment, Position of the
  driver, etc.
• Occupant information is the same as the driver
  with the exception of Condition.

8
Mississippi Uniform Crash ReportVehicle

• Includes information about the vehicle involved
  in the crash.
• Fields included are Sequence of Events, Vehicle
  Action, Vehicle Configuration, Insurance
  Information, Owner of Vehicle information,
  Direction of travel, etc.
• Also, includes Commercial Vehicle information.
  This information is required based on the Vehicle
  Configuration selection.

9
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report

• Officer takes completed form to the police
  office

10
Officer Takes Completed Form to the Police Office

• Reporting officer carries
• the completed report to
• his / her police station or
• Highway Patrol Troop and
• gives to the supervising
• officer.

11
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office

• Reviewing officer ensures data on form is accurate

12
Reviewing Officer Ensures Data on Form Is Accurate

• Process of review puts a second set of eyes on
  the report to catch any errors that may have been
  overlooked by the submitting officer.
• Objective of recording accurate data is to ensure
  that the crash can be entirely recreated from the
  data captured.

13
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is accurate

• Forms are sent to be scanned

14
Forms Are Sent to Be Scanned

• Reporting agency sends the completed report to
  the Mississippi Department of Public Safety,
  Safety Responsibility Division.
• Safety Responsibility Division is responsible for
  scanning the forms into the system,
  electronically pulling data from the scanned
  forms, and data verification.
• Department of Public Safety Troop Secretaries
  actually scan the forms in-house and are
  responsible for the data verification process
  themselves.

15
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is
  accurate
• Forms are sent to be scanned

• Forms are scanned, checked for scan quality

16
Forms Are Scanned, Checked for Scan Quality

• Operator receiving the crash report is
  responsible for the scan quality, page
  orientation, and page order.
• Operators were trained on the equipment,
  software, and the techniques to ensure quality.
• Page order is a manual verification of the image
  it is imperative that Page 01, with the unique
  Agency Number and Agency Case Number, is the
  first page scanned of the crash report.
• Order of the subsequent pages does not affect the
  electronic data pull.

17
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is
  accurate
• Forms are sent to be scanned
• Forms are scanned, checked for scan quality

• Teleform Scanning Station

18
Teleform Scanning Station

• Teleform Scanning Station electronically creates
  an image file of the crash report, in a .tif
  format
• System creates and assigns batches of crash
  reports for electronically pulling of data from
  the images and data verification
• Automated process and does not require human
  intervention

19
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is
  accurate
• Forms are sent to be scanned
• Forms are scanned, checked for scan quality
• Teleform Scanning Station

• Verification process

20
Verification Process

• Teleform Reader software then uses Optical
  Character Recognition (OCR) technology to read
  the scanned information.
• Automated process that pulls data from the
  scanned forms and creates an electronic version.

21
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is
  accurate
• Forms are sent to be scanned
• Forms are scanned, checked for scan quality
• Teleform Scanning Station
• Verification process

• Verifier reviews scanned information

22
Verifier Reviews Scanned Information

• Verifier reviews the electronic copy of the form
  to ensure that accurate data was captured.
• Individual must then correct data where there
  were Optical Character Recognition (OCR) /
  scanning errors.

23
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is
  accurate
• Forms are sent to be scanned
• Forms are scanned, checked for scan quality
• Teleform Scanning Station
• Verification process
• Verifier reviews scanned information

• Teleform Verifier exports data to the staging
  database

24
Teleform Verifier Exports Data to the Staging
Database

• Staging database is the last checkpoint prior to
  be entered into the production database.
• During the export, there is an initial data
  validation check, and errors identified are
  published in the error tables.

25
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is
  accurate
• Forms are sent to be scanned
• Forms are scanned, checked for scan quality
• Teleform Scanning Station
• Verification process
• Verifier reviews scanned information
• Teleform Verifier exports data to the staging
  database

• Electronic data validation

26
Electronic Data Validation

• After all data has been entered into the staging
  database, more stringent electronic validation
  applies greater constraints to the data before it
  is moved over to the production database.

27
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is
  accurate
• Forms are sent to be scanned
• Forms are scanned, checked for scan quality
• Teleform Scanning Station
• Verification process
• Verifier reviews scanned information
• Teleform Verifier exports data to the staging
  database
• Electronic data validation

• Return of error data for correction and
  reprocessing

28
Return of Error Data for Correction and
Reprocessing

• Electronic process pulls data from the error
  database and uses a publishing agent to recreate
  the Mississippi Uniform Crash Report for problem
  reports.
• The form can be emailed or physically printed and
  sent to the originating agency for review and
  correction.

29
Mississippi Uniform Crash Report Optical
Character Recognition Process Description

• The officer on the scene fills in the MS Uniform
  Crash Report
• Officer takes completed form to the police office
• Reviewing officer ensures data on form is
  accurate
• Forms are sent to be scanned
• Forms are scanned, checked for scan quality
• Teleform Scanning Station
• Verification process
• Verifier reviews scanned information
• Teleform Verifier exports data to the staging
  database
• Electronic data validation
• Return of error data for correction and
  reprocessing

• Data in production database

30
Data in Production Database

• Reports that are in the staging database, which
  pass all verification checks, the data is ported
  over to the production database.
• Data is now ready for reporting and analysis.

31
Overall Pitfalls

• Dirty scanner bed or dirty documents
• Optical Character Recognition (OCR) software has
  a hard time recognizing characters that were
  printed too heavily, so that they bleed together,
  or too lightly, so that the thinner portions of
  the characters are broken.

32
Pitfalls of the Officer Filling Out the Report on
the Scene

• Reporting Officers handwriting was a huge
  factor in Optical Character Recognition (OCR)
  Scanning because every officers handwriting is
  unique.

33
Pitfalls of Reviewing Officer Ensuring Data on
Form Is Accurate

• Over 100 business rules attached to the fields
  on the different forms. Impossible for Reviewing
  Officer to catch every mistake on each and every
  report that was reviewed.

34
Pitfalls for Forms Are Sent to Be Scanned

• Safety Responsibility
• (SR) Division would
• receive large amounts
• of reports in the mail
• from law enforcement
• agencies on a daily or
• weekly basis.

35
Pitfalls of Forms Are Scanned, Checked for Scan
Quality

• Safety Responsibility Division (SR) supervisor
  would separate reports into batches and assign to
  one of three persons that were in charge of
  scanning reports.

36
Pitfalls of Verifier Reviews Scanned Information

• Optical Character Recognition (OCR)
• recognition was so poor that the operator
• had to basically re-key the entire report.

37
Pitfalls on Teleform Verifier Exporting Data to
Staging Database

• Reports were verified and the master table
  verification stored procedure ran to check for
  errors between related forms, this would generate
  literally thousands of errors. After several
  months of error accumulation, it was impossible
  for the operators to see the light at the end of
  the tunnel.

38
Solutions to Optical Character Recognition (OCR)
Scanning Problems

• Traffic Records Committee began seeking and end
  to this nightmare resulting from Optical
  Character Recognition (OCR) Scanning.
• Optical Character Recognition (OCR) Scanning
  process had started in January 2004 and a Request
  for Proposal (RFP) for a new automated crash
  report system went out shortly after that date.
• Bid was awarded to Visual Statement, Inc. for its
  electronic collision reporting software
  ReportBeam in late June of 2004.
• Department of Public Safety (DPS) began working
  with Visual Statement to customize the ReportBeam
  software for the State of Mississippis specific
  needs.

39
Solutions Continued

• Several law enforcement agencies outside the
  Department of Public Safety umbrella were chosen
  to participate in a pilot of ReportBeam in
  October of 2004.
• Once the testing phase with the pilot agencies
  was completed, the DECISION was made to have
  Safety Responsibility Division (SR) to start
  using ReportBeam to enter the crash reports that
  were submitted to the Department of Public Safety
  (DPS) for data entry since they were already
  manually entering the entire report with the OCR
  system anyway.
• At least with ReportBeam, the errors could be
  corrected at the time of entry through the
  extensive run-time business rules instead of
  having to be rescanned and correcting the reports
  several times before the data was entered into
  the Department of Public Safety database.
• To this day, the initial data captured by the OCR
  system is being scrubbed for import into the
  States ReportBeam data repository.

40
Example of Error at Time of Entry From ReportBeam
41
Solutions Conclusion

• Once ReportBeam was officially accepted by the
  Commissioner of Department of Public Safety,
  training began for all law enforcement agencies
  in February of 2005 and continues even today.
• Once a law enforcement agency is trained, their
  officers will enter their own crash reports into
  the system and submit them directly to the state
  server via the Internet.
• State of Mississippi has approximately 300 Law
  Enforcement Agencies and we currently have over
  200 Law Enforcement Agencies using ReportBeam.
• Department of Public Safety is very happy to see
  the number of Agencies committed to using an
  Automated Crash Reporting System like ReportBeam.

42
Questions ???