Investigation into Travel-Time Route Interpolation and Prediction, 10-R9708

Printer Friendly Version

Principal Investigators
Robert W. Heller
Brandon Meiners

Inclusive Dates:  04/01/07 – 09/01/08

Background - Intelligent transportation systems (ITS) provide information to motorists about traffic conditions in the form of travel times to destinations and aberrant roadway conditions. Systems publish travel times based on sensors reporting current traffic conditions, failing to track evolving traffic conditions and possible sensor data trends. System performance degrades when sensors do not provide data, resulting in less accurate travel times or no travel times at all. Travel times are calculated based on past traffic conditions, i.e., reported speeds of traffic as it moved past a sensor. When a motorist reaches the beginning of a route and a travel time is provided, that travel time is already obsolete; travel times are calculated oblivious to events that have occurred after the sensors reported their data.

Approach - The travel time route interpolation and prediction (TRIP) system investigated perceived natural associations between traffic sensor data, the relationships between data points in the route, and variations over time by capturing trends in the data. The "standard" against which improvements were compared was calculating travel times based on current sensor data. This method is used commonly in Advanced Traffic Management Systems (ATMS), is documented widely in the industry literature, and SwRI staff members have implemented this method in several systems, and made use of historical data provided from a production ATMS system in the analyses.

Accomplishments - This project explored various techniques to synthesize missing sensor data from traffic detection systems, finding that simple linear spatial interpolation provided better results when compared to quadratic or cubic spatial interpolation or temporal extrapolation; however, better results are obtained when historical data is available that meets existing traffic conditions. Further, the project explored techniques to relate travel time computations using current detector data to travel time calculations utilizing "aged" data finding inconclusive results under steady state conditions (free-flow and congestion), but very promising results mining historical data during transition from free flow to congestion and the reverse.

2008 Program Home