1. Hossein Naraghi Traffic Records Forum August 7, 2017
Institute for Transportation (InTrans)
Observing and modeling drivers’ natural driving behavior in work zones
Hossein Naraghi Traffic Records Forum August 7, 2017

2. Background 11% increase in work zone fatalities (2011 – 2015)
Presence of road constructions and maintenance
Increase in safety risks
Disturbance to traffic flow
High cognitive work load for drivers
fatalities per year in last 10 years
11% increase in work zone fatalities (2011 – 2015)
Slight decrease in non-work zone fatalities
Limited details on available crash data
Officer interpretation, Witness testimony
Unclear driver behavior and underlying causes of work zone crashes
Institute for Transportation (InTrans)

3. Background Speed management effectiveness
Numerous countermeasures tends to get drivers attention to reduce speed
How drivers react to measures like DMS, merge sign,..?
How effective are safety strategies in getting driver attention?
The most effective countermeasure cannot be identified because driver behavior is not well understood
SHRP2 NDS Time Series speed data collected at high frequency at 10HZ (every 0.1 second)
Opportunity for a first hand observation of driver behavior
Institute for Transportation (InTrans)

4. Background Major contributing factors Speeding Inattentive driving
Driver
The more recent study by NHTSA in 2015 confirms numbers and reveals critical reasons attributed to
Drivers %
Vehicles %
Environment 2%
Unknown %
Limitation in understanding human factor in available crash data
Roadway
Vehicle
Rumar et al.’ study (1985)
NHTSA Traffic safety Facts (Data Source: NMVCCS 2005–2007)
Institute for Transportation (InTrans)

5. Objectives Utilize SHRP2 NDS time series data and forward video images
Observe drivers behavior
Determine where drivers starts to react to different work zone features
Identify the effectiveness of speed countermeasures
Develop models to predict driver reactions to work zone features
Institute for Transportation (InTrans)

6. Primary Participant Count
Data collection
SHRP 2 Naturalistic Driving Study Data (NDS)
3100 participant drivers
Oct 2010 – Nov 2013
6 Sites
5 M trips
20 M data miles
4 M GIGs data
Site
Primary Participant Count
Buffalo, NY
719
Tampa, FL
698
Seattle, WA
676
Durham, NC
504
Bloomington, IN
239
State College, PA
256
VTTI, 2013
Institute for Transportation (InTrans)

7. Data collection SHRP 2 Roadway Information Database (RID)
25,000miles of roadway data
Curve Radius
Lane Types
Curve Length
Shoulder Types
Curve Point of curvature
All Signs on MUTCD
Curve Point of tangency
Guardrails/Barriers
Curve Direction
Intersection Geometry
Grade
Median Types
Cross-slope/Super elevation
Presence of Rumble strips
Number of Lanes
Presence of Lighting
Lane Width
A site map shows the RID database in the SHRP2 NDS (Smadi, 2015)
Institute for Transportation (InTrans)

8. Methodology Functional data Analysis (FDA)
Institute for Transportation (InTrans)

9. Data extraction and reduction
FDA analyses
Major contribution in locating WZ within 511 data in RID Database and matching to WZ identified in SHRP2 NDS data
State specific attribute query in ArcGIS with key word search in “Event Description” field
Construction, Lane closure, Road work, Maintenance
WZ for less than 3 days duration excluded
Low possibility of having NDS trip
9290 potential WZ identified
Link the WZ to NDS trip density map
Potential WZ determined if trip falling within dates indicated in 511
Institute for Transportation (InTrans)

10. Data extraction and reduction
4800 time series traces along with forward/rear video
Roadway characteristics
# of lanes, median type, traffic control, speed limit, …
WZ characteristics
DMS, lane closure, speed feedback sign, WZ speed limit, lane shift, shoulder closure, ….
Drive characteristics
Age, gender, and other socioeconomic data provided by VTTI
Countermeasure distance
Calculated based on timestamp and speed
Institute for Transportation (InTrans)

11. Data quality assurance
Forward video image quality
Excluded traces with very poor image quality
Unable to identify the position of vehicle
Unable to confirm active WZ
Speed time series data issue was missing data
Excluded traces with more than 25% missing values
Utilize interpolation techniques to calculate missing values
Institute for Transportation (InTrans)

12. Chall
enge
Institute for Transportation (InTrans)

13. Challenges of time series data
Use of time series data brought many challenges
Missing data and noises
Outliers
Lack of well-established statistical method
Institute for Transportation (InTrans)

14. Functional Data Analysis (FDA)
FDA emerged in early 90s and quickly developed more recently in medical field, environmental monitoring and economic research
In FDA a series of discrete time series observations converted to functional observations
Primary interest is to understand the variation in underlying process over a group of repeated measurements
Appropriate for high frequency measurement, smooth process, examine complex process and explain how process changes over time

15. Functional Data Analysis (FDA)
Development of FDA from discrete time series data
First step smooth noises
Values can evaluated at any points from best fitted function
Institute for Transportation (InTrans)

16. Smoothing splines 4-lane divided Rural Left lane closure Cross-over
head to head traffic
Best fitted spline to each trace
62 traces
Institute for Transportation (InTrans)

17. Functional Data Analysis (FDA)
Best fitted spline to each curve along with discrete observations
Institute for Transportation (InTrans)

18. Functional Data Analysis (FDA)
Scree plot to identify number of basis functions
Generalized cross validation used to choose optimal smoothing parameters
Institute for Transportation (InTrans)

19. Functional Data Analysis (FDA)
Polynomial segments with order of four polynomials
17 spline basis functions defined over the interval (-1.3, 1.0) by 13 interior knots
Institute for Transportation (InTrans)

20. Functional Data Analysis (FDA)
62 best fitted spline functions based on 17 spline basis
Institute for Transportation (InTrans)

21. Functional Data Analysis (FDA)
Institute for Transportation (InTrans)

22. Solid black line is average speed profile
Useful technique to summarize average behavior from a group of time series data
Institute for Transportation (InTrans)

23. Institute for Transportation (InTrans)
26 Traces closed lane
26 Traces closed lane
62 Traces All
33 Traces female drivers
36 Traces open lane
29Traces male drivers
Institute for Transportation (InTrans)

24. Institute for Transportation (InTrans)

25. First derivative
First derivative of vehicle speed indicates vehicle acceleration
Useful to technique to illustrate how drivers change their speeds when encountering various countermeasures
Institute for Transportation (InTrans)

26. First derivative
Institute for Transportation (InTrans)

27. First derivative Institute for Transportation (InTrans) 62 Traces All
26 Traces closed lane
62 Traces All
33 Traces female drivers
36 Traces open lane
29 Traces male drivers
Institute for Transportation (InTrans)

28. Conclusions Functional Data Analysis (FDA)
FDA method very useful and effective in analyzing time series data
The typical driver’s behavior could be visualized and calculated with FDA methods
The derivative information could provide important insights on time series data
It is more important to study the rate of change
This method allow us to examine how a group of drivers
reacted to different WZ characteristics or speed
countermeasures
Institute for Transportation (InTrans)

29. Conclusions
Identify which work zone features are more likely to get driver’s attention
DMS vs. to static sign
DSFS vs. WZ speed limit sign
Determine the effectiveness of the locations of the countermeasures
DMS located after 1st warning sign
DMS after the merge sign
Investigate the effectiveness of WZ configurations
left lane vs. right lane closure
lane shift no shoulder vs. lane closure
Institute for Transportation (InTrans)

30. Future Research
Utilize Multiple Changepoint and FDA analysis to study driver behavior
Alternative locations of countermeasures
WZ configurations
Nighttime versus daytime comparison
Driver demographics
Distractions’ effects
Institute for Transportation (InTrans)

31. All things are difficult before they become easy?
Have patience
All things are difficult before they become easy ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?
Thank YOU ? ? ? ?
Institute for Transportation (InTrans)