Estimating origin-destination traffic flows from the random variability in automatic detector counts : a preliminary analysis.

From time to time, it is necessary to estimate traffic movements between individual origins and destinations on a road network. The results are essential for modelling the effects of traffic management and road improvement schemes. However, conventional origin-destination (OD) survey methods are very expensive. Recently, a number of alternative methods have been proposed in which estimates of the OD matrix are obtained from traffic counts. They require as input either an old OD matrix, or a traffic model. However, a much simpler approach may be feasible. The number of vehicles making a journey from A to B on a network varies randomly from moment to moment and from day to day. These variations contain potentially useful information; under appropriate conditions, the observer may be able to link variations in traffic flow at one location with variations at another location with sufficient precision to determine the volume of traffic flowing between them. Hence, it may be possible to estimate OD matrices from traffic counts on an empirical basis, without any other information. Some methods have already been suggested in which solutions are obtained by recursive methods applied to systems of simultaneous linear equations. However, in general this approach requires information about all the flows entering and leaving the study area network, and this may be inconvenient when dealing with an urban centre extending over several blocks. Here, a much simpler alternative is proposed, which can be applied to the estimation of a single 'through' traffic stream if required. In addition, with this approach it is possible to take account of the variation in journey times of vehicles between each origin and destination, and to estimate the distribution of journey times as well as the OD flows themselves. The method involves establishing a statistical relationship between the flows observed at the origin and those observed at the destination, either in terms of correlation coefficients, or via a straightforward linear regression. There seems to be no reason why the estimates should not be generated on-line, if data is available from permanent detectors installed as part of a computer-controlled traffic signal system, such as SCOOT (Split Cycle and Offset Optimisation Technique). In this paper, a number of possible procedures for doing this is suggested, and a preliminary assessment is made of their performance.