Data Collection Through New On- The-Road Technologies: The Need for Validation David Harkey University of North Carolina Highway Safety Research Center.

Data Collection Through New On-The-Road Technologies: The Need for Validation

David HarkeyUniversity of North Carolina

Highway Safety Research Center

Traffic Records ForumJuly 2003

Denver, Colorado

UNC Highway Safety Research Center Highway Safety Information System

Overview of Presentation

Roadway Inventory: Needs and Methods

Major Missing Element: Alignment History of Recent HSIS Efforts Results from New HSIS Study Conclusions


Roadway Inventories

Inventory databases exists in all State DOTs

Used for planning, operations and maintenance

Range of data elementsgeometrics (roadways, intersections and

interchanges)traffic control/operations (segment speed

limits, intersection TCD, signal timings)


Roadway Inventory Collection

Manual methodsIn-field acquisition (surveys, observational

runs with DMI or GPS technology) In-office acquisition (video logs, aerial

photos, satellite images) Automated Methods

Instrumented Vans (video logs, pavement conditions, sign inventory, geometrics)


Alignment Data

Very few states have curvature and grade data

Only 2 of the 9 HSIS states have good horizontal curvature data

Known relationship between curvature and safety


HSIS Research Focus

Improve upon the alignment data available for safety analyses

Automated methods using instrumented vehicles ARAN - Automatic Roadway

Analyzer

Roadware Group, Inc.


History of HSIS Efforts

1996/97 - MN Data Collection and Analysis4 2-lane roads with 50+ curves over 21

milesPoor Results - Consistency and Accuracy

1999 - Re-analysis of MN Data with algorithm developed for CT DOTDramatic improvement in consistency and

accuracy


New HSIS Effort

Evaluate the CT DOT algorithm that uses ARAN data

Repeatability (Consistency) between multiple runs on the same

segment of roadway Accuracy - Comparison to “ground

truth”As determined from ground surveys


Data Collection Tasks

Site Selectionapproximately 50 miles distributed by

number of lanes and level of curvature ARAN Data Collection

5 passes in each direction on every routeazimuth data acquired every 4 meters

Survey AlignmentsCT DOT task from ground surveys

segments


Data Processing

Horizontal Curve Classification and Display System (PLV-HC Software)Develop alignments from raw data

Make post-processing adjustmentsRemove unresolved (unknown) segments at

the end of a runConvert spirals to tangents and arcs (split the

difference in length)Remove “true” outliers (not be more than 1 in

the group of 5 runs)


PLV-HC Software


PLV-HC Software


PLV-HC Software


Visual Confirmation

Video log produced for each run

Allowed for the removal of induced curves


Example Analysis (Rte 190)

Initial Best-Fit AlignmentLog Analysis 1

Log Analysis 2

Log Analysis 3

Log Validation 1

Log Validation 1

Rev Analysis 1

Rev Analysis 2_QC

QC_Harkey

QC_Van Dine

Rev Validation 1

Rev Validation 2

Rev Analysis 3



Adjustment 1 - removal of unknowns (unresolved

segments)Log Analysis 1

Log Analysis 2

Log Analysis 3

Log Validation 1

Log Validation 1

Rev Analysis 1

Rev Analysis 2_QC

QC_Harkey

QC_Van Dine

Rev Validation 1

Rev Validation 2

Rev Analysis 3



Adjustment 2 - Conversion of Spirals to Arcs and Tangents

Log Analysis 1

Log Analysis 2

Log Analysis 3

Log Validation 1

Log Validation 1

Rev Analysis 1

Rev Analysis 2_QC

QC_Harkey

QC_Van Dine

Rev Validation 1

Rev Validation 2

Rev Analysis 3



Kappa Statistics for Log Direction

0.64 0.91 0.91 3 Analyses

0.51 0.83 0.89 2 Validations

0.66 0.87 0.91 3 Analyses + 2 Validations

Kappa Statistics for Reverse Direction

0.83 0.85 0.87 3 Analyses

0.72 0.89 0.84 2 Validations

0.79 0.89 0.89 3 Analyses + 2 Validations

0.91 0.91 0.89 Quality Control Analysis (H&Y)

0.91 0.93 0.92 Quality Control Analysis (H&Y&D)


Consistency Kappa Statistics

11 sites (22 sets of results - log and rev)

K > 0.90 9 sets 0.90 > K > 0.75 8 sets K < 0.75 5 sets



Accuracy AssessmentLog Analysis 1

Log Analysis 2

Log Analysis 3

Log Validation 1

Log Validation 1

Survey

K = 0.92


Accuracy Kappa Statistics

11 sites (log direction only)

K > 0.90 2 sites 0.90 > K > 0.75 4 sites K < 0.75 5 sites


Conclusions

Are the results consistent?77 percent of the directional groups met the

threshold

Are the results accurate?55 percent of the sites met the threshold

Neither consistent nor accurate enough


Conclusions

Need to conduct individual arc analysisvariation in length and radius across

multiple runsvariation in PCs and PTsdifferences with survey results

Critical to validate the results of any automated data acquisition technology


For the Latest Info

Visit the HSIS web site

www.hsisinfo.org

Data Collection Through New On- The-Road Technologies: The Need for Validation David Harkey University of North Carolina Highway Safety Research Center.

Documents

safety slide

rev analysis

tangents log analysis

analysis s4

rev validation

log validation

accuracy slide

colorado slide