Transnow Student Conference February 9, Techniques for Mining Truck Data to Improve Freight Operations and Planning Zachary Horowitz Portland State University Transnow Student Conference Oregon State University, Corvallis, Oregon
Transnow Student Conference February 9, Presentation outline Introduction Project Background Sampling Methodology Data Collection Results Conclusions Next Steps Acknowledgements
Transnow Student Conference February 9, Why collect freight data? Economic importance Lack of knowledge of freight transportation system operation Safety and security Positive affects on overall transportation system
Transnow Student Conference February 9, Project background Basic idea: Record freeway traffic in order to compare truck counting methods 3 methods: 1) Manual 2) Autoscope “speed” and “count” detectors 3) Nihan-Wang algorithm (UW) “Team Autoscope” infamous around ITS Lab
Transnow Student Conference February 9, Data sources PORTAL Portland area traffic data archive Collects volume & occupancy data at 20 second intervals ODOT surveillance CCTV network About 80 cameras on local highways Recorded to DVD Used for both manual counts and Autoscope Autoscope Recorded video data processed using RackVision and accompanying software
Transnow Student Conference February 9, Site selection 3 locations in Portland, OR I-5 SB at Marine Dr. I-84 WB at Sandy Blvd/37 th I-5 NB at Lower Boones Fy. 10 days over 4 weeks Traffic recorded primarily From noon to 5 pm
Transnow Student Conference February 9, Camera positioning Autoscope sensitive to poor viewing angles Camera height Field of view Poor light / direct sunlight Camera must be able to view detector area Pan / tilt / zoom (PTZ) functionality Manual counts used as “Ground truth” ODOT TMOC staff assisted with positioning Limited choice in site locations
Transnow Student Conference February 9, Count station setup Fiber optic cable feed from ODOT TMOC with 70+ “channels’ of traffic
Transnow Student Conference February 9, Manual data collection Recorded traffic viewed with DVD player Timer.exe used to count both long (>39 feet) and short (< 39 feet) vehicles in shoulder and median lane Data exported to Excel for processing Volumes determined Cumulative (N,t) curves made
Transnow Student Conference February 9, Nihan-Wang algorithm Dual-loop speed trap functionality from single loop Volume and occupancy data Vehicle length estimation PORTAL comparison tab Algorithm returns: Count of long / short vehicles for each 20 second time period Speed measurements for Nihan-Wang and ODOT reported
Transnow Student Conference February 9, Video image processing Establish connection with Communication Server Set RackVision parameters Detector Editor Calibration Speed detectors Count detectors Save file and upload to RackVision
Transnow Student Conference February 9, Creating and running data polls Create data collection poll Add detectors to poll Play video and start poll Vehicles pass over detectors Presence (count) Speed and length Iterative calibration process Data is stored on RackVision in Flash memory Download when hour is complete and export (.txt file) to Excel
Transnow Student Conference February 9, Data organization and processing Manual count data sorted by: Lane Vehicle type (short / long / combined) Cumulative and oblique plots Histogram of Autoscope speed detector data used for length determination in some cases
Transnow Student Conference February 9, Data Analysis Summary statistics table For each count method: Total vehicles Total SV Total LV Speed measurements % differences Across lanes Across sites
Transnow Student Conference February 9, Graphical plots Cumulative count of all vehicles Oblique graph of all vehicles plotted using a scaling factor Oblique graph of short vehicles using a scaling factor
Transnow Student Conference February 9, Comparison data Comparison of total count mechanisms to ground truth Truck count percent difference
Transnow Student Conference February 9, Comparison data Measured average speed relation to all count error Comparison by site
Transnow Student Conference February 9, Conclusions Nihan-Wang estimations Unusually large error for some speeds and counts Detector issues such as loop sensitivities Incorrect or missing data Autoscope Difficulties in calibration and data processing Returned favorable results compared to ground truth ODOT surveillance camera system not ideal for Autoscope Manual counts still produce best data
Transnow Student Conference February 9, Next steps Additional data collection Detector fidelity is key Different sites may produce better data Vehicle classification by type Further statistical testing PORTAL data calibration will improve Nihan-Wang algorithm implementation results Travel time estimation for freight vehicles Permanent, automated freight vehicle counting Ramp meter speed trap installations
Transnow Student Conference February 9, Acknowledgements ODOT TMOC Staff Dr. Robert Bertini PORTAL Team: Dr. Kristin Tufte Spicer Matthews Jonathan Horowitz
Transnow Student Conference February 9, References [1] Y. Wang and N. Nihan. Dynamic estimation of freeway large truck volume based on single-loop measurements. CD-Rom for the 80th Annual Meeting of TRB, paper , TRB, National Research Council, Washington D.C., [2] Kwon, Jaimyoung. Joint Estimation of the Traffic Speed and Mean Vehicle Length From Single-Loop Detector Data. CD- Rom for the 82nd Annual Meeting of TRB, National Research Council, Washington, D.C [3] Kwon, Jaimyoung; Varaiya, P. P.; Skabardonis, Alexander. ESTIMATION OF TRUCK TRAFFIC VOLUME FROM SINGLE LOOP DETECTOR USING LANE-TO-LANE SPEED CORRELATION. CD-Rom for the 82nd Annual Meeting of TRB, National Research Council, Washington, D.C [4] Nihan, NL; Wang, Y; Zhang, XP. EVALUATION OF DUAL-LOOP DATA ACCURACY USING VIDEO GROUND TRUTH DATA. TransNow, Transportation Northwest, Washington Univ, Civil Engineering Dept, [5] Y. Wang and N. Nihan. Can Single-Loop Detectors Do the Work of Dual-Loop Detectors? ASCE Journal of Transportation Engineering, 129(2), , 2003 [6] National ITS ADUS Addendum [7] S. Turner. Guidelines for Developing ITS Data Archiving Systems. Report FHWA, U.S. Department of Transportation, Texas Department of Transportation and Texas Transportation Institute, [8] Y. Wang and N. Nihan. A Robust Method of Filtering Single-Loop Data for Improved Speed Estimation. CD-Rom for the 81st Annual Meeting of TRB, paper , TRB, National Research Council, Washington, D.C [9] P. Athol. Interdependence of Certain Operation Characteristics within a Moving Traffic Stream. Highway Research Record 72,
Transnow Student Conference February 9, Questions and Comments