Transcript OR-RLC-Presentation-5-18-08

Red-Light Cameras
The Good
The Bad &
The Uncertain
Dale Gedcke, B.Eng., M.Sc., Ph.D.
Marketing & Technical Consultant
Oak Ridge, TN
OR RLC Presentation 5-18-08.ppt
How RLCs Work
Crossing sensor on red
captures ±6 sec of video
Traffic
Light
Sensor
Car
Camera
If you cross the
sensor on red,
about 2 weeks later
a citation like this
arrives in the mail
for the owner of the
vehicle.
The black bars include the
time, place, speed limit,
your speed, and amount of
time the light was red when
your crossed the sensor.
Where to view the
video on-line.
Red-Light Violation Response Options

View video on-line, then chose one of 3 options:
1) Pay $50

No points on license
No notification to insurance company

No notification to TN Dept. of Safety

2) Name the actual driver



Affidavit
Vehicle owner is liable if actual driver fails to respond
No notification to Insurance Company, nor TN Dept. of Safety
3) Contest the ticket in court. If you lose:


Additional $8 scheduling fee, plus $60 court fee
Judgement becomes a public record available to data mining companies
Knoxville, TN / Redflex Citations in 2007
• Revenue Sharing Formula:
< $4500 per camera per month: 15% to City, 85% to Redflex
> $4500 per camera per month: 50% to City, 50% to Redflex
• 15 Intersections with cameras
• 60,299 Red-light violations
• $955,014 to City revenue
• $1,644,719 to Redflex revenue
• Net = $43.11 per initial citation. (Compare to $50 ticket.)
RLC Safety Premise
Installing Red-Light Cameras should improve
safety because:



Drivers running a red light risk a collision with crosstraffic operating on green
Red-Light Cameras catch ALL red-light violations
Drivers soon learn to stop on red to avoid a ticket
Why Do Drivers Run Red Lights?


In a rush. Tried to beat the red. (Sensitive to RLC)
Misjudged time versus distance (Minor sensitivity to RLC)



Did not see the signal (Not sensitive to RLC)


Visibility problem
Could not stop in time (Not sensitive to RLC)


Thought there was time to make it on yellow
Distracted when it changed to yellow
Dilemma Zone (yellow too short for speed limit)
CONCLUSION: RLC will not suppress all
red-light running.
Proving/Disproving the RLC Safety Premise


Simple analyses are almost always misleading
Requires a comprehensive and sound statistical analysis because:






Many confounding variables (traffic rate, traffic patterns, weather,
intersection characteristics, speed, time of day, day of week, changes in
vehicle safety features, driver characteristics, truck traffic, yellow duration,
all-red duration, etc.)
Must correct for changes in traffic volume
Must compare the before and after accident rates to similar intersections
without RLCs (to determine what changes were due to the RLCs)
Important to test for spill-over effects at nearby intersections without RLCs
Paucity of data and poor quality of data
Dealing with low numbers of random accidents: Statistical uncertainty is
large.


To detect a 10% difference with 95% confidence requires >800 accidents
Before/After RLC installation and >1600 Before/After at non-RLC control sites.
Must use tests for statistical significance of trends


t-test, Chi-squared test, F-test, variance analysis, regression analysis,
empirical Bayes analysis
If an analysis does not use any of these tests, its conclusions are worthless.
Several Sound Statistical Studies are Available

Barbara Langland-Orban et al., Red Light Running Cameras: Would Crashes,
Injuries and Automobile Insurance Rates Increase If They Are Used in Florida?
Florida Public Health Review, 2008; 5:1-7, and 5: 47-52.

Nattaporn Yaungyai, Evaluation Update of Red Light Camera Programming in
Fairfax County, Virginia; Master’s Thesis, Virginia Polytechnic Institute and State
University, April 2004.

Mark Burkey and Kofi Obeng, A Detailed Investigation of Crash Risk Reduction
Resulting from Red Light Cameras in Small Urban Areas, North Carolina
Agricultural & Technical State University, Greensboro, NC, July 2004.

Nicholas J. Garber et al., The Impact of Red Light Cameras (Photo-Red
Enforcement) on Crashes in Virginia, Virginia Transportation Research Council,
Charlottesville, VA, June 2007.
Barbara Langland-Orban et al., Florida, 2008

Excellent review and critique of all statistical studies
performed on RLC installations to date.

Easy to read for the non-statistician.

A good place to start!
Nattaporn Yaungyai, Thesis, Virginia Polytechnic Institute
and State University, April 2004.











Excellent introduction to the RLC technology, history, laws and
statistical testing process
Summarizes laws in all USA States extant up to April 2004.
Summarizes experience in several countries
Reviews public opinion surveys on RLCs in various cities.
Analyzes RLC impact in Fairfax County, VA
13 camera intersections; 12 to 82 accidents per year per intersection
category (RLC, comparison, spill-over); only 2 comparison
intersections; 4453 to 887 total RLC violations per month
Inadequate number of accidents and too few comparison
intersections for determining significant accident trends.
Enough RLC violations to determine RLC violation trend.
RLCs reduced red-light violations as much as 58% in 22nd 27th month after installation
Lengthening yellow light duration reduced red-light violations
as much as 70%
Statistically non-significant reduction in accident rate.
Burkey and Obeng, Greensboro, NC, July 2004.

Large data base yields good statistical accuracy:









303 intersections
18 RLC sites: 840 accidents before & 777 accidents after RLC installation
285 non-RLC sites: 4827 accidents before, 4211 after
Analyzed correlation with weather
RLC installation associated with a statistically significant 40%
increase in accident rates compared to non-RLC intersections.
No change in angle accident rates, and large increases in rear-end
crash rates and other types of crashes relative to non-RCL
intersections
Statistically non-significant increase in fatal accident rate
Decrease in left-turn accident rate with cross traffic at RLC sites
Longer yellow light duration decreases accident rate
Garber et al., VA Trans. Res. Council, 2007

Large data set yields good statistical accuracy in the aggregate:




More than 3500 crashes over 7 years
28 RLC intersections and 44 non-RCL intersections
6 jurisdictions (VA): Alexandria, Arlington, Fairfax City, Fairfax County, Falls
Church & Vienna
Analyzes aggregate versus individual jurisdictions and intersection
types for differing trends. (Statistical accuracy suffers when subdivided by
jurisdiction or intersection.)







Offers guide for selecting successful accident-reduction RLC sites
Attempts cost vs. benefit analysis: Inconclusive +/- result
Aggregate total accident rates increased 29% with RLC
Aggregate Rear-end crash rates increased 42% with RLC
Aggregate red-light running crash rates decreased a statistically
non-significant 8% with RLC
4 of RLC intersections experienced increased angle crash rates
2 of RLC intersections showed decreased rear-end collision rates.
Dilemma Zone
You can get trapped by a
dilemma zone, if the
yellow duration is too
short for the speed limit.
Traffic
Light
Sensor
dc
Camera
ds
Dilemma Zone:
If light turns yellow, it is
impossible to enter the
intersection before red,
and impossible to stop
safely before the
intersection.
Dilemma Zone
Approach speed = v = 40 mph = 58.7 ft/s
Yellow light duration = tY = 4.0 seconds
Decision time = tD = 1.00 seconds
Reaction time = tR = 0.50 seconds
Maximum safe deceleration rate = a = 10 ft/s/s
Percent Grade (+ve uphill, -ve downhill) = G% = 0
Maximum distance to make it on yellow:
d c  v tY  235 ft
Minimum distance to stop safely (avoiding rear-end collision):
d s  vt D
v2
 t R 
 260 ft
2a  0.32G% 
Dilemma Zone exists from 235 to 260 ft from intersection.
If the light turns yellow when you are in this zone you can neither continue safely
nor stop safely.
Minimum Yellow to Eliminate Dilemma Zone
From 2003 TN Traffic Design Manual,
recommended yellow light durations for level intersections are:
Approach Speed (mph): 25 30 35 40 45 50 55 60 65
Yellow Duration (sec.): 4.0 4.0 4.0 4.5 4.5 5.0 5.0 5.5 6.0
NOTE: Descending grades require longer delays.
Using values from the previous slide, the minimum yellow duration at
40 mph is computed as:
tY min  t D  t R  
v
 4.4 sec.
2 a  0.32 G% 
For more realistic 85th percentile reaction times (tD + tR), see: 1) APPENDIX B and 2) Thomas J. Triggs and
Walter G. Harris, Reaction Time of Drivers to Road Stimuli, Human Factors Report No. HFR-12, ISBN 0 86746
147 0. Human Factors Group, Department of Psychology, Monash University, Victoria 3800, Australia,
June 1982.
Conclusions

The most important and most effective safety solution: Increase
yellow duration according to approach speed and grade to eliminate
unsafe Dilemma Zones.

Statistically sound studies show that Red-Light Cameras are NOT
a reliably effective solution for improving safety.





Red-Light Cameras generate addictively large revenues



The RLC effect on angle collision rates varies from positive to negative
RLCs usually significantly increase the rate of rear-end collisions
RLCs tend to increase total accident rates
However, RLC citation rates do decline in the first two years of operation
Conflict of Interest: Safety can easily become less important than
revenue
Analyze the operating costs carefully. Some municipalities have lost
money because of increased direct labor, overhead, and court costs.
Some have discontinued RLCs due to declining revenue.
See Appendices A through D for more details.
Appendix A: Useful References





US DOT Federal Highway Administration Manual on Uniform Traffic Control Devices
(MUTCD): http://mutcd.fhwa.dot.gov/
Tennessee MUTCD Traffic Design Manual:
http://mutcd.fhwa.dot.gov/resources/state_info/tennessee/tn.htm
Making Intersections Safer: A Toolbox of Engineering Countermeasures to Reduce
Red Light Running, FHWA and Institute of Transportation Engineers, Pub. No. IR115, ISBN: 0-935403-76-0 (2003):
http://safety.fhwa.dot.gov/intersections/rlr_report/index.htm (published before the
2004 and 2007 studies on RLCs)
Dr. Peter T. Martin, Vikram C. Kalyani and Aleksander Stefanovic, Evaluation of
Advanced Warning Signals on High Speed Signalized Intersections, Univ. of Utah,
Nov. 2003.: http://www.mountain-plains.org/pubs/html/mpc-03-155/index.php
(Dilemma Zones)
Philip R. Bevington and D. Keith Robinson, Data Reduction and Error Analysis for the
Physical Sciences, WCB/McGraw-Hill, ISBN 0-07-911243-9, 1992. (Excellent
introductory book on statistical analysis)
Appendix B: Driver Reaction Times
Dependent on several factors:

Distractions
Anticipation of, or focus on a particular warning or hazard
Difficulty of interpreting the significance of the hazard or warning
Difficulty of figuring out the best solution
Physical motion required to implement the solution
Mechanical delay in vehicle actuation

What is the Right Reaction Time?








MUTCD documents use a 1-second reaction time with a yellow light.
Dilemma Zone example used 1.5 seconds.
Studies showed a longer yellow reduces accidents and red-light
violations.
Appendix B (cont’d.): Driver Reaction Times.
1.5 sec. is a reasonable minimum
Driver Reaction Time to Surprise Encounter: Triggs and Harris 1982
Around a blind curve, past the crest of a hill, or triggered ON at the last moment
Mean
Standard
85th
Reaction Deviation Percentile
Time (s)
(s)
(s)
Field Test
Car Following (response to lead car brake light)
0.92
0.28
1.26
Protruding Vehicle with Tyre Change
0.97
0.45
1.50
Lit Vehicle Under Repair at Night
1.02
0.36
1.50
Flashing Railway Crossing - Night (Rally Drivers)
1.14
0.34
1.50
Flashing Railway Crossing - Night (General Population)
1.18
0.36
1.50
Flashing Railway Crossing - Day
1.77
0.84
2.53
Speed Trap: Tynong
1.75
0.7
2.54
Parked Police Vehicle (just past crest in hill)
2.37
0.69
2.80
C.R.B. "Roadworks Ahead" Sign
1.64
1.26
3.00
Speed Trap: Beaconsfield
2.46
1.04
3.40
Speed Trap: Dandenong North
2.45
0.92
3.60
Speed Trap: Gisborne
2.54
0.66
3.60
Traffic Engineers typically use the 85th Percentile for design:
i.e., 85% of drivers have a lower reaction time.
From: Thomas J. Triggs and Walter G. Harris, Reaction Time of Drivers to Road Stimuli, Human Factors Report No. HFR-12,
ISBN 0 86746 147 0. Human Factors Group, Department of Psychology, Monash University, Victoria 3800, Australia, June 1982.
Appendix C: Required Yellow Duration and Green Delay
to Eliminate Dilemma Zone and Clear the Intersection
Maximum Deceleration Measurement & Basic Parameters: Enter Measured Red Values
Time to recognize
Minimum Time to
Effective
change from Green Time to Move foot Duration of
Initial
Stop, after
Minimum
Deceleration to Yellow light and from Accelerator
Yellow
Speed applying brake
Stopping
Constant
decide whether or
to Brake pedal
Light
(mph)
(sec.)
Distance (ft.)
(ft/sec/sec)
not to stop (sec.)
(sec.)
(sec.)
40
5.86
171.9
-10.0
1.0
0.5
4
Minimum Yellow Duration to Eliminate Dilemma Zone
Distance
Minimum
Travelled (ft.)
Distance
Required
Initial
while
travelled (ft.) Distance Travelled
Yellow
Speed
Initial Speed
deciding to while moving while Decelerating Total Distance to Duration
(mph)
(ft/sec)
stop
foot to brake
(ft.)
Stop (ft.)
(sec)
5
7.3
7.3
3.7
2.69
13.7
1.9
10
14.7
14.7
7.3
10.74
32.7
2.2
15
22.0
22.0
11.0
24.17
57.2
2.6
20
29.3
29.3
14.7
42.97
87.0
3.0
25
36.7
36.7
18.3
67.15
122.1
3.3
30
44.0
44.0
22.0
96.69
162.7
3.7
35
51.3
51.3
25.7
131.61
208.6
4.1
40
58.7
58.7
29.3
171.89
259.9
4.4
45
66.0
66.0
33.0
217.55
316.6
4.8
50
73.3
73.3
36.7
268.58
378.6
5.2
55
80.7
80.7
40.3
324.99
446.0
5.5
60
88.0
88.0
44.0
386.76
518.8
5.9
65
95.3
95.3
47.7
453.91
596.9
6.3
70
102.7
102.7
51.3
526.42
680.4
6.6
75
110.0
110.0
55.0
604.31
769.3
7.0
Green Delay is also
called the “All-Red”
phase (red for all
directions).
Required Green Delay vs. Intersection Width
40-ft
60-ft
80-ft
100-ft
120-ft
Width:
Width:
Width:
Width:
Width:
Green
Green
Green
Green
Green
Delay
Delay
Delay
Delay
Delay
(sec)
(sec)
(sec)
(sec)
(sec)
5.5
8.2
10.9
13.6
16.4
2.7
4.1
5.5
6.8
8.2
1.8
2.7
3.6
4.5
5.5
1.4
2.0
2.7
3.4
4.1
1.1
1.6
2.2
2.7
3.3
0.9
1.4
1.8
2.3
2.7
0.8
1.2
1.6
1.9
2.3
0.7
1.0
1.4
1.7
2.0
0.6
0.9
1.2
1.5
1.8
0.5
0.8
1.1
1.4
1.6
0.5
0.7
1.0
1.2
1.5
0.5
0.7
0.9
1.1
1.4
0.4
0.6
0.8
1.0
1.3
0.4
0.6
0.8
1.0
1.2
0.4
0.5
0.7
0.9
1.1
Appendix D: Poisson Statistics.
Why the Sample Size Must Be Large
 = the mean or average of the
distribution.
A measure of the width or
dispersion of the distribution about
the mean is the standard deviation, s
s 
Poisson Probability Distribution for Mean = 9
0.14
0.12
0.10
P(N)
• Accidents have a low probability:
typically < 50 per 1 million cars thru
the intersection.
• Poisson Statistics apply: the
probability of observing N accidents
in a time period t is:
 N eN
P( N ) 
N!
0.08
Mean
0.06
s
0.04
s
0.02
0.00
0
5
10
15
N
20
25
Appendix D (cont’d.): Poisson Statistics.
Why the Sample Size Must Be Large
For a single sample of N accidents observed in a time t, the best estimate of the mean of
the underlying probability distribution is
N
The best estimate of the standard deviation of the underlying probability distribution is
s  
N
The dispersion, expressed as a percent of the mean value is
s%
s 100%
N

100%
N
s% represents the percent uncertainty in estimating the true, mean
number of accidents from a single measurement, N.
Appendix D (cont’d.): Poisson Statistics.
Why the Sample Size Must Be Large
Percent Uncertainty in the Number of Accidents
N:
1
10
100
1,000
10,000
s% :
100
32
10
3.2
1
100,000 1,000,000
0.32
0.1
• To detect changes of the order of a few percent, one must have
a sample of more than 10,000 accidents.
• When comparing accident rates at RLC intersections to similar,
non-RLC intersections, one needs more than twice as many non-RLC
intersections to keep from degrading the statistical uncertainty.