Advanced Traffic Surveillance Using Vehicle Signatures

1
Advanced Traffic Surveillance Using Vehicle
Signatures

• Stephen G. Ritchie
• Department of Civil and Environmental Engineering
  andInstitute of Transportation
  StudiesUniversity of California, Irvine

.
2
Outline

• Objectives
• Benefits
• Field data collection
• Results to date
• Ongoing research

3
Research Partners

• Research team of academic, private and public
  sector partners led by UC Irvine, in
  collaboration with
• GTS Inc (and Professor Reinhart Kuhne of
  University of Stuttgart), 3M Inc, Peek Traffic
  Inc (US/UK), Caltrans and the City of Irvine
• Contract manager R. Tam, PATH
• Technical Monitor J. Palen, Caltrans

4
Objectives

• develop and implement intelligent traffic
  surveillance system
• anonymous vehicle tracking based on vehicle
  signatures from non-intrusive sensors
• initially utilize conventional inductive loops
• derive real-time section-related traffic
  performance measures for ATMIS
• travel time, speed, density, vehicle class, O-D
  information

5
Potential Real-Time Applications

• Traveler information
• Congestion monitoring
• System performance
• Incident detection
• Ramp control
• Vehicle classification
• Dynamic traffic O-D estimation
• Dynamic traffic assignment
• Loop condition assessment

6
Some Key Components

• Conventional inductive loops
• high-speed scanning detector cards
• 2070 field implementation
• Smart software to reidentify vehicles,
  including
• tailgating, vehicles with trailers, turning
  vehicles, off-center vehicles, lane switching
  over loops, other irregular signatures

7
Field Data Collection 1

• 1.2 mile, 4-lane section of SR-24 freeway in
  Lafayette, California
• 14 hours over 6 days in December, 1996
• Peek detectors 12 millisecond scan rate
• Ground truthing video data collected
  simultaneously

8
Feature Vectors

• Waveform shape
• Maximum amplitude
• Speed
• Electronic vehicle length
• Lane number

9
SR-24 Reidentification Results

• Matching car - 75, station wagon - 75, SUV -
  100, Pickup - 80, truck - 61, van - 92,
  semi-trailer - 100
• Congested flow (1800 vphpl)
• density - 3.6average error (46.5 based on
  occupancy)
• travel time - 3.2 average error (45.8 based
  single speed trap)

10
Initial Vehicle Classification Study Results (1)

• SR-24 freeway signature data, approx. 1000vphpl
• Three heuristic discriminant algorithms explored
• Seven vehicle classes used for a sample of
  vehicles

11
Initial Vehicle Classification Study Results (2)

• Overall classification rate (one heuristic on
  test data) - 85
• Vehicle classes used
• 1. car, minivan, station wagon - 96
• 2. SUV, pickup - 72
• 3. van, full-size pickup - 77
• 4. limo - 67
• 5. 2 axle truck - 75
• 6. vehicle trailer, bus - 100
• 7. 3 or more axletruck - 75

12
Field Data Collection 2

• Alton Parkway arterial in Irvine, California
• 425 foot 2-lane section with double loops
• analysis based on 2.5 hours of data from AM and
  Noon periods on June 30, 1998
• approx. 600-800vph total, for both periods
• 3M detector cards, approx. 8ms scan rate
• ground truthing video data collected
  simultaneously

13
Signature Analysis (1)

• Alton Parkway, AM data, front loops
• 604 vehicles
• video-based irregularities
• heavy vehicles
• left turning vehicles
• off-center vehicles
• lane switching
• tailgating
• motorcycles

14
Signature Analysis (2)

• Preprocessing step required for SR-24 algorithms
  for
• noisy, bi-modal and odd-shaped signatures
• left turn and off-center vehicles
• vehicles with trailers (count as 1, not 2 vehs)
• tailgating (with full overlap of signatures in
  record count as 2 vehs, not 1)
• tailgating with records containing only the
  leading edge of the second vehicle

15
Signature Analysis (3)

• Alton Parkway, AM data, front loops
• Signature-based analysis
• 594 records from data logging software
• 597 vehicles after signature preprocessing (3
  full overlap tailgaters)
• plus 7 undetected vehicles/signatures (3
  motorcycles, and 4 lane changes by small vehicles
  between the loop stations)
• Total 604 vehicles
• Volume count accuracy 597/604 98.8 (99.3,
  if motorcycles were detected)

16
Alton Parkway Preliminary Reidentification
Results (1)

• Based on interim parameter calibration of SR-24
  algorithm
• 597 vehicles less 5 bad downstream signatures
  (all partial overlap tailgaters) less 7 upstream
  signatures missing at t0, yields 585 vehicle
  population with signatures up and downstream

17
Alton Parkway Preliminary Reidentification
Results (2)

• Initial reidentification match 80.0
• Matches by class
• passenger car/wagon(419) 77.3
• SUV (71) 90.1
• pickup (47) 89.4
• truck (3) 100.0
• van (6) 83.3
• minivan (38) 78.9
• bus (1) 0

18
Alton Parkway Single Loop Speed Estimation

• Hypothesis vehicle speed is proportional to
  gradient or slew rate of waveform leading edge
• Vehicle speed vs slew rate appears linear
• AM R0.89, N581 Noon R0.78, N530
• Linear regression on AM data (R20.83, N300)
  yielded average speed error of 6.7
• Conventional single loop method (assumed
  effective vehicle length) gave errors of about 37

19
Prototype Field Implementation

• Major 4-way signalized intersection in City of
  Irvine
• Tasks
• advanced algorithm development and testing
• system and communication architecture design, for
  a loop-data collection system with transmission
  of processed data to the Irvine TMC
• 2070 software development and testing
• field installation, shakedown and operation

20
Intersection Performance

• Real-time data at the TMC for both operators and
  applications can include
• approach and exit volumes (and headways), by lane
  and total
• turn movement volumes
• approach and exit speeds
• travel times through the intersection, by
  movement
• vehicle class
• Enables determination of
• LOS and intersection performance
• congestion and incident detection
• intersection O-D analysis

21
Real-Time Intersection LOS

• Highway Capacity Manual (HCM) uses Control Delay
  for LOS
• Irvine intersection, loop placement and
  reidentification of vehicles can provide a
  similar (but different) measure
• Initial study based on video data under light to
  moderate traffic (LOS A and B) showed promising
  correlation between the reidentification-based
  and HCM delays.
• Will be investigated further in a follow-on
  project

22
Concluding Comments

• Accurate section-based measures of traffic
  performance are obtainable from conventional
  loops with this approach
• Potential exists for cost-effective network-wide
  management, especially with the current extensive
  loop infrastructure
• Speeds can be obtained from single loops with
  reasonable accuracy
• Promising results also obtained for vehicle
  classification and real-time level of service
  estimation
• Prototype field implementation underway in Irvine