On-line estimation of traffic split parameters based on lane counts

Abstract

Adaptive traffic control systems need to continuously monitor traffic conditions and predict immediate traffic conditions to respond to changes in both recurrent and nonrecurrent traffic patterns. One measure of traffic conditions is the number of turning vehicles at intersections and freeway offramps. Split parameters may be estimated from the numbers of turning vehicles, and along with upstream traffic counts, these may be used to predict the numbers of downstream vehicle arrivals. This study develops several responsive methods to estimate split parameters for four-legged intersections and freeway segments from vehicle counts. For intersections, these methods depend on the geometric layout of the intersection and the signal stage. The vehicle counts are collected by signal stages and lanes. The split parameters or turning proportions are estimated for each signal stage and, then they are combined to estimate the turning proportions for the entire interval. Some methods need counts for only one cycle to estimate turning proportions while others need additional data. For those methods that need additional data, four closed-form estimation methods are developed. Two methods need prior turning proportion estimates: (1) maximum entropy (ME) and (2) generalized least-squared error (GLS). The other methods require counts for three cycles: (3) least-squared error (LS) and (4) least-squared error/generalized least-squared error (LS/GLS). Results from these methods are compared with each other. The ME, GLS and LS methods which use cycle counts are also developed and their results are compared with the results of their stage-count counterparts. For freeway segments, a virtual box method for consistent vehicle counting is developed. Three split parameter estimation methods are developed for freeway segments. One method, GLS, uses counts, from one virtual box and requires prior split parameter estimates. The other methods, LS and GLS/LS, need several virtual boxes, depending on the number of interchanges in the freeway segment. Split parameter estimation approaches are also developed for small road networks by combining split parameters from individual intersections and freeway segments.