Test Intersection Status Results Preparation for State Data

Test Intersection: Status, Results, Preparation for State Data Collection Lee Alexander Pi-Ming Cheng Alec Gorjestani Arvind Menon Craig Shankwitz Intelligent Vehicles Lab University of Minnesota

Presentation u u System Overview Test Intersection Status v v u Examples of Data Collected v v u u Construction Sensing Data collection Analysis Animations Video Database status Installation in partner states v v Design Documents Cost

System Overview u Mainline surveillance v v u Minor Road Surveillance v v u Laser based system provides “profile” of stopped vehicle Used for data analysis, timing of warnings (when DII deployed) Crossroads Surveillance v v u Radar based sensing Provides position, speed, lane assignment, and time to intersection of each sensed vehicle Used to capture driver behavior (one step or two) NOT part of deployed IDS system Computation v v Acquires driver behavior data now Compute warning timing when IDS deployed

Construction - Mainline

Construction - Mainline

Construction- Vehicle Classification

Construction – Crossroads Surveillance

Construction- Control Cabinet

Test Intersection Status u Mainline surveillance Construction Complete v All sensors operational v Series of Validation Experiments Complete v

Mainline System Performance Results u Detection Rate: 99. 990% (5 “misses” out of 51, 942) v v Miss defined as vehicle not within 40 meters of test zone Results for a single sensor; multiple sensors decrease likelihood of a “miss”

Mainline System Performance Results u Lane Data Accuracy v v Longitudinal Accuracy 8 meters Lane assignment accuracy 90% • Ambiguity during lane changes, hanging near center line • Limitations of angular resolution of radar v Speed accurate to 1 MPH

Mainline System Performance Results u Vehicle Shadowing Performance: Range accuracy will be no worse than 75 feet if lateral shadowing occurs Performance: Can resolve 2 vehicles if separated by 50 or more feet

Vehicle Classification Validation Configuration DGPS Camera Horizontal Laser Vertical Laser Presence Detector Vehicle Classifying Radar

Vehicle Classification Performance u Accuracy approximately 85% based on vehicle height v One sensor reduces cost substantially u Grouping actual conservative…. classify as larger than

Vehicle Classification: Height Only Cars Lt. Truck SUV Med. Truck Semi. Truck

Crossroads Surveillance u Positions based on locating front of vehicle v v v u Performance v v u Working definition of gap Accuracy 1 -2 meters Larger concern 1 step or 2, time in crossroads Left turns sensed, captured correctly 95% of time Right turns sensed, captured correctly 95% of time Open Issue v v Straight through captured only 60% right now Camera issues • Absolute vs. Relative thresholds (being tested today/tonight) • IR Illuminators Cheaper ($2 k system vs. $26 K system) We control illumination

Crossroads Trajectory Tracker Validation: Day with Visible Light Camera

Crossroads Trajectory Tracker Validation: Night with IR Camera

Crossroads Trajectory Tracker Validation: Night with IR Camera

Intersection Surveillance System: Visualization of all Data

Data Collection: Visualization

Data Acquisition – Control Cabinet

Data Acquisition – IV Lab Data Flow/Archival 120 Gbyte IDE Drive requires replacement once every 2 weeks. DOT will have to dispatch someone to swap out to mail to U of MN.

Batch program finds vehicles entering intersection from minor road (Vehicles of Interest (VOI) ) and consolidates tracking information to new table User specified queries User specified results Data Acquisition – IV Lab Analysis Processes

Data Acquisition and Analysis u Database system has been design u Initial automated queries have been completed. u Will be validating results next two weeks u Automated and specialized queries supported

Automated queries (can be run as frequently as desired). u Gaps as a function of vehicle classification u Gaps as a function of time of day u Gaps for Right, Straight, Left turns u Percentages of one step vs. two step maneuver u Identification of near misses/accidents u Other queries supported as well. v Add license plate reader, further refine data set.

Data Analysis – cont’d u Weird v Observational Data For every 100 Right turns, • 100 Straight-throughs • 5 left turns v 2 drivers have missed intersection approaching from west, none have missed from right • Both crashes, damages could have been much worse. • Last crash, no sensor damage, just mount damage

Can we build one for you?

FLASHBACK! Intersection Build Details from AP 2004 • Radar Stations • Vehicle Classification Stations • Vision Systems • Central Cabinet • Ethernet and Video Cable $191, 837

Cost Data u Electrical Contractor: $101 K v v v Bought guys lunch last week in Cannon Falls, must be happy Rethinking Laser for Vehicle Classification Two Crash repairs: • #1, $2500 • #2, $800

MN vs. Partner States Cost u Minnesota Partner States Radar Subsystem: $50 K Video (Xroads): $70 K Minor Roads: $48 K Cabling (Power and Data): $10 k v Grand total (Contractor + HW): $314 K v v v u u v Radar Subsystem: $50 K Video (Xroads): $35 K • 2 masts, not 4 • SDRC with IR Illumination, not IR Cam. v v u Minor Roads: $35 K (2 lasers, not 4) Cabling (Power and Data): $10 k Computer Assy, parts procurement: $20 K Estimated total (Contractor+HW): $275 K

Build one for you? u Final Reports due 28 Feb 2005 u Master Design Document will be appendix u Can make available to states who wish to review/help them make decision to install equipment.