Automatic Vehicle Location (AVL) systems are increasingly being introduced across bus networks around the world to the benefit of passengers and operators. The performance of these systems is dependent on the quality of input data, one of the most important being the accuracy of the bus-stop locations. This paper outlines how GPS measurements collected when the bus opens its doors can be used to validate the location of bus-stops. London Buses intend to apply such an approach to validate the location of all its 19,000 bus-stops in London. Three approaches are tried, two based on univariate density estimation and one based on multivariate density estimation. In all cases a Kernel estimator is used. Data collected from a bus route in London fitted with the new iBus AVL system is used to determine which approach is best. The results suggest that a univariate kernel density estimate with a correction factor to allow for the stop being at the side of the road performs best.
Abstract