Estimation of Route Travel Time Distribution with Information Fusion from Automatic Number Plate Recognition Data

Route travel time varies with vehicles and traffic demand. Besides the average route travel time, route travel time reliability in the form of travel time distribution is indispensable. However, the sample size of Complete Route Travel Times (TTC) is rather small for many reasons. Existing methods using convolution distribution rely on strong assumptions about the correlation structure or the link travel time distributions; other methods relying on scaled Partial Route Travel Times (TTP) may extend the estimation bias. To overcome these issues, we present an estimation method for route travel time distribution by fusing kinds of route travel time information from Automatic Number Plate Recognition (ANPR) data. The proposed method firstly improves the data quality for estimation in four steps, including route redefinition, observation extraction, path inference, and scaling. Secondly, using TTP data, it convolutes the empirical travel time distributions on all the partial routes divided at the breakpoints identified by the Hopkins statistics. Thus, the link correlations are considered and the assumption about the correlation structure is eschewed. Thirdly, the convolution distribution and TTC information are fused to estimate the actual route travel time distribution based on Bayes' theorem and Shannon's information entropy. Finally, estimation results using different methods are compared to evaluate the developed model.