Improving Travel Time Estimates from Inductive Loop and Toll Collection Data with Dempster–Shafer Data Fusion

Abstract

cannot associate a 100% probability of certainty to the events the data represent. The technique captures and combines whatever certainty or knowledge exists in the event classification capability of the information sources. The knowledge contributed by the information sources is combined, by using Dempster's rule, to find the conjunction of the events and the associated probability or belief that the decision is correct (1-4).The U.S. Department of Defense (DoD) data fusion model and Dempster-Shafer inference, in particular, as a tool for combining multisource data are discussed in the following sections. Field test data from inductive loop detector (ILD) pairs and toll collection stations are then used to demonstrate application of the technique to travel time analysis and estimation. This process involves computing a confusion matrix that describes the likelihood the sensors or toll collection devices are reporting correct travel times. Information derived from electronic toll collection (ETC) with tags, real-time credit card payments from the credit-card-only lane, and cash payments are used to find the "true" travel time values (used as the reference value) against which travel time estimates gathered from ILD data, toll tag data only, and toll tag plus real-time credit card payment data are compared. Travel times computed from toll tag data alone are an important travel time information source as those data can be supplied in real time to traffic management centers and used to supplement travel times derived from loops. DATA FUSION MODELMany data-processing techniques originally developed by the DoD to support identification and tracking of military objects, including Dempster-Shafer inference, can be used today to aid traffic management on surface streets and highways (5-7). The DoD data fusion model consists of a hierarchy of five processing levels. Level 0 deals with preprocessing of data from the contributing source. It may normalize, format, order, batch, and compress input data (7,8). It may even identify subobjects or features in the data that are used later in Level 1 processing.For highway and arterial traffic management, Level 1 processing concerns data gathering from all appropriate sources, including realtime point and wide-area traffic flow sensors, transit system operators, toll data, cellular telephone calls, emergency call box reports, probe vehicle and roving tow truck messages, commercial vehicle transmissions, and roadway-based weather sensors (3).Level 2 processing identifies the probable situation causing the observed data and events by combining the results of Level 1 processing with information from other sources and databases. These Availability, accuracy, and reliability of systemwide travel time information contribute to effective decision making in support of safe and efficient operation of traffic management systems. Such information is made available in modern traffic management systems through sensor and data-processing technologies that gather real-time traffic and weather da...

Data fusion in intelligent transportation systems: Progress and challenges – A survey

“…Results display propensity of proposed schemes for estimation accuracy improvement. More recently, the evidence theory was used to solve the same problem [17,23]. In these contributions, travel time was broken into classes and formulate the estimation problem as a classification one.…”

“…El Faouzi et al [7,8,17] proposed an estimation framework for real-time traffic characterisation based on multi-source data. As an illustrative example, a multi-source travel time estimation was performed based on two data sources: data from conventional loop detectors which deliver Eulerian data and probe vehicles collecting Lagrangian data.…”

“…-Probabilistic approaches: for instance, Bayesian approach with Bayesian network and state-space models [10], maximum likelihood methods and Kalman filter based DF [11,12], possibility theory [13], evidential reasoning and more specifically evidence theory [14][15][16] are widely used for the multi-sensor data fusion. This later technique could be viewed as a generalization of Bayesian approach [15][16][17]. -Artificial intelligence: neural networks and artificial cognition including artificial intelligence, genetic algorithms and neural networks.…”

How much GPS data do we need?

“…One day of GPS-based floating car data are used as the validation set. Another study uses Dempster-Shafer inference to classify travel times into one of four categories using loop and toll collection data on a motorway in France (Faouzi et al, 2009).…”

Estimating a large drive time matrix between ZIP codes in the United States: A differential sampling approach

“…One way to account for traffic effect in drive time estimation relies on the utilization of traffic sensors or auxiliary data sources. These sensors include loop detectors (Kwon et al, 2000;Coifman, 2002) and automatic vehicle identification systems-such as toll collection system (El Faouzi et al, 2009), license plate recognition system (van Hinsbergen et al, 2009), and Bluetooth-based system (Bhaskar and Chung, 2013)-that are installed at certain locations along the road. They can accurately capture travel speeds and times.…”