Traffic monitoring during extreme [ News Gazette , 12] congestion - - PDF document

6/7/13 1

1

Traffic monitoring during extreme congestion events Dan Work

Assistant Professor, Civil and Environmental Engineering & Coordinated Science Laboratory University of Illinois at Urbana–Champaign

[News Gazette, ‘12] 2 2

Limitations of current systems

• Surface streets

– Sparsity of sensing – Limited (but increasing) GPS data from mobile devices

• Rely on statistical

algorithms

– Heavily influenced by historical priors

6/7/13 2

3 3

Extreme congestion events

• Event driven congestion

– Sporting events – Natural disasters

• Impact on transportation

infrastructure

– Network topology changes – Damage to physical components – Loss of cyber components – Change in travel demands Need for cheap, instantly

deployable (temporary) sensing

[A. Savulich, New York Daily News, 2012]

4 4

TrafficTurk smartphone app When a vehicle passes the intersection, swipe its movement on the screen.

6/7/13 3

5 5

Inspiration for TrafficTurk

Amazon’s Mechanical Turk Turning movement counters (Transportation’s Mechanical Turk) The mechanical Turk

6 6

100+ sensors deployed to monitor football traffic 220,000+ vehicles swiped 140 volunteers

6/7/13 4

7 7 8 8

TrafficTurk Experiment - NYC

• Hurricane Sandy –

November 3 and 4, 2012

• Garment District, Manhattan
• Overnight map deployment
• Recruitment at Columbia

University

• Real disaster response

experience

10+ hours monitoring

[NSF RAPID # 1308842] [Scientific American Citizen Science featured project ‘12]

6/7/13 5

9 9

• Goal: identify traffic signal

phases from maneuver data

• Motivation:

– Recovery of traffic phase timings – Simplified TrafficTurk user interface

Processing techniques: Phase Inference via Hidden Markov Modeling

100 200 300 400 500 600 Phase1 Phase2 Phase3 Phase4 Phase5 Phase6 Time(sec) Phase real estimated

[M. Reisi Gahrooei & D. Work, IEEE ITSC 13]

10 10

Processing techniques: Inverse optimal traffic signal control

• Goal: recover traffic signal control

logic via learning on the cost function

• Motivation:

– Flow model forecasting on surface streets – Limited information on existing infrastructure (none at large scales) – Human traffic control

[S. Gowrishankar & D. Work, IEEE ITSC 13]

6/7/13 6

11 11

• Processing TrafficTurk data for NY (phase detection

controller detection)

• Integration into real-time traffic estimation algorithms
• Acquiring (FOIL) NYC GPS taxi data pre and post

Sandy.

Next steps

12

Traffic monitoring during extreme congestion events Dan Work

Assistant Professor, Civil and Environmental Engineering & Coordinated Science Laboratory University of Illinois at Urbana–Champaign

[News Gazette, ‘12]