Transcript Status update - ITS Institute

Rural Intersection Decision Support
(IDS) System: Status and Future Work
Alec Gorjestani
Arvind Menon
Pi-Ming Cheng
Lee Alexander
Bryan Newstrom
Craig Shankwitz
University of Minnesota
ITS Institute
Intelligent Vehicles Lab
Presentation Overview





Present status
Validation/Characterization work
Optimization work
Data collection
Data analysis





driver behavior
4 seasons of data, 24/7
Additional technical capabilities for CICAS
Future Work
Anecdotes
Present Status




All Systems working (showed yesterday)
Open Architecture, we can integrate most any sensor,
communication system, processor, etc.
Mostly off-the-shelf hardware
Need to add

Wireless communication
• Add hardware at radar station cabinets
• Add hardware at main controller cabinet

Radar based vehicle classification
• IV Lab has sensors
• No J1708 message set for vehicle classification capability
• Eaton-Vorad reorganizing, point of contact difficult to find
Present Status (cont’d)

Need to add

Delphi Mainline Radar
•
•
•
•

Purchase Order June 2004
3 Sensors ordered for comparison to Vorad
Not yet arrived
Calls to Delphi not returned.
Would really like DSRC Test Kit
(wink wink, nudge nudge)
Validation work: vehicle masking
sensitivity

Radar sensitivity analysis
“masking” of small radar Xsections by large ones
 Distance at which a motorcycle is masked by a
passenger car or truck

X
Use DGPS equipped
probe vehicles to determine
X at which motorcycle is
masked
Validation work: radar
detection/accuracy validation

Radar detection miss
rate





Use two reference sensors
as a measure against
radar detection and range
accuracy
Light beam known location
Broken beam triggers
interrupt
Compare radar data with
light beam
(presence/location) and
camera
(presence/location)
Complements previous
accuracy work
Radar
Camera
DGPS-Probe Vehicle
Vehicle/Gap Tracker
Tracker program Kalman filter-based state
estimator
 Noisy radar signals

Internal sensor processing tends to “pull in”
vehicle position along sensor longitudinal axis
 If uncompensated, leads to lane assignment
errors (azimuth errors)

States include vehicle location, vehicle
speed, vehicle heading, lane assignment
 Gap tracking = 1-vehicle tracking

Validation work: Vehicle/Gap Tracker
Performance



All vehicles entering intersxn (both major and minor
roads) assigned ID
All vehicles tracked within intersxn boundaries
DGPS position compared to tracker position/ID for
lane changes, left turns, right turns, speed variations,
etc.
ID 144
DGPS-Probe
Vehicle
ID 225
ID 169
Validation work: Vehicle/Gap Tracker
sensitivity to loss of radar


Possibility radar may fail
Tracker program designed to



Detect radar loss
Compensate for radar sensor loss
Validate by disabling radar, running program, and
comparing DGPS-based state estimate with tracker
estimate
ID 144
DGPS-Probe
Vehicle
ID 225
ID 169
Validation work: Vehicle Classification
System performance

Compare radar & laser based system performance






$1200 system vs. $13,000 system
Determine performance envelope for Benefit:Cost analysis
Presence verified by light beam sensor
Reference is visible light and IR Cameras aimed at
minor roads
Image processing results compared to radar and
Lidar results. If three agree, performance is as
expected. (Automation improves efficiency)
Discrepancies analyzed by human viewing captured
images


Identify problem areas
Improve system capability
Vehicle Classification Validation
Configuration
Crossroads Trajectory Tracker
Validation
DGPS-Probe Vehicle
Camera
Optimization work: Radar

Radar Sensor Spacing
Intersection overbuilt
 Presently, 100% coverage
 Each sensor, 400’ range
 Tracker good enough for 500’ spacing? 600’
spacing?

• Less spatial density => lower sensor cost
Less trenching
Lower power
Lower maintenance
Lower cost
Optimization Work: Radar

Radar Sensors Considered
Presently, Eaton
 Delphi Ordered

• Will be installed as soon as they arrive
• Specifications close to Eaton Vorad
• Considerably more expensive

Autosense? Decision based on VTTI’s results.
• Autosense specs close to Eaton, but much longer range
• Considerably more expensive than Eaton.
• Geometric considerations – “seeing” over a hill

CA COTS Study Promising Technology
Optimization work: Communication

Wired vs. Wireless communication
Original thought was to go wireless. However, given
effort to trench power, wired communication was an
incremental cost.
 Wireless

• Pros:
Offers significant cost savings: i.e., no trenching
• Cons:
Unknown reliability, sensitivity to local EMI conditions
Sufficient bandwidth for present and future
applications?
Optimization work: Communication
Wired
Pros:
•Known bandwidth, known reliability, immunity from local EMI
Cons:
•Trenching costs, wire breakage, etc. (incremental cost not too
great if power trenching done at same time).
Hardwired
DSL to outside world for analysis,
diagnostics, streaming video
Data Collection

Sensors

Mainline Radar
• Location, speed, heading, lane

Xroads: Camera, Laser, Radar
• Vehicle position, heading

Minor road Laser, Radar
• Vehicle length, height profile


Remote Weather Information System 0.9 miles North of
Intersxn
Rates




Most sensors at 10 Hz
Laser at 30 Hz locally, processed data at 10 Hz
Video at 30 Hz
Weather at 15 minute intervals
Data Collection

Formats



Engineering data stored as a database of “snapshots” of the state
of the intersection at 10 Hz
Video data .mpg4 at 30 Hz. 5 Cameras
4 Gbyte data/day
 Storage
 Local 80 Gbyte removable drive
 2 Terabyte server at the U
Data Collection

Access



DSL at the Intersection, monitor status remotely
Mn/DOT truck station streaming video (maintenance, response)
Quality Assurance



Data checks
Periodic back-ups
Self-diagnostics
Data Analysis

Understand Driver Behavior


Statistics (Howard Preston’s work) showed that far-side (left
turn) crashes (70% in general, 80% at our intersection) far
outweigh nearside crashes
WHY?
• Right turns “easier?”
• Drivers take left in one motion rather than 2 (pause in median)?

Distribution of gaps accepted by drivers: what gaps are being
taken?
•
•
•
•
for right turns
for left turns
for crossing intersection
How Safe are these?
Data Analysis – cont’d

Correlate driver behavior with


Vehicle type / size (vehicle classification)
Driver age (macroscopic level, limited basis initially license
plate reader later)
• Limited basis means grad student observer





Driver gender (limited basis initially, license plate reader later)
Weather effects (R/WIS 0.9 Mile away), with in-road sensors
(collecting data already)
Plan to collect data for 12 months, and analyze
incrementally
Results directly applicable to Human Interface final
design/deployment and algorithm strategy
Provides baseline measure for Field Operational Test
Future work

State Pooled Fund study underway



Portable Surveillance system


7 states included
Goal is to instrument intersection in each state, determine
regional behavioral differences with drivers
Sensors and comm. system built to analyze rural intersections
(upcoming proposal to Minnesota Local Road Research Board)
Add microscopic driver data

License plate reader can yield driver age/gender information
• Important to understand crash causality


May eventually allow “tailoring” of warnings to specific driver
Early analysis complete. Details to be worked out
• Data from DPS
• Analysis
MN IDS Intersxn CICAS Capability Communication

Wireless communication

Presently
• 2.4 GHZ 802.11B
 Range about 1.6 Km
• 900 MHz RF Modem
 Range about 4-6 Km

Future
• Mesh Networks
• DSRC (5.9 GHz)
• Emerging Technology
 4 Foldable masts
 4 transmit/receive sites
 Easy to change HW
 Not tied to a particular
architecture
MN IDS Intersxn CICAS Capability Communication

Differential GPS corrections


Architecture Analysis





Intersection validation, mapping
Data broadcasts
Client/Server
Router/switch
Bandwidth needs testing / analysis
Intersection state information comm.


Collision avoidance
Communication of data/in-vehicle warnings
MN IDS Intersxn CICAS Capability Communication

Map Downloads
Map detail (we have layers of
detail/info)
 Range – how much /fundamental
details needed for the intersection
 Timing (data well in advance of the
xroads)
 Handshaking/verification

• Validation that vehicles which need data
have it
MN IDS Intersxn CICAS Capability Sensors

Differential GPS corrections




Road Weather sensors


Warnings / notifications to vehicles
Vehicles as sensors



Methodology
Correction source
Validate accuracy requirements
Road friction
Position/speed/heading for collision avoidance
Other

If it plugs in, we can use it.
Anecdotes

Local residents in favor of this technology
“dangerous intersection!”
 “this will be great.”
 “will that slow traffic on 52?”
 “will that issue tickets?”
 “when are you going online?”
 “last winter, a LOT of cars went into the ditch…”


Two crashes have occurred since construction
began in May

Right angle crash resulted in injuries (stretcher and
ambulance)
This is a good summer job.