| Probe based traffic monitoring involves monitoring a subset of vehicles in the traffic stream at different locations in space and time as they traverse the roadway. Since probe vehicles represent samples in a traffic stream, probe based traffic monitoring, at its core, relies on effective sampling of vehicles. Current research on sampling has not fully addressed all issues relevant to the sampling problem. The development of a sampling design involves several steps the most important of which are (1) determination of required sample sizes and (2) choice of sampling method for effective distribution of samples. Most research in sampling has focused on the former aspect but little has been published on the choice of sampling method.;In sampling for probe based traffic monitoring, the current practice is to specify a single sample size as a percentage of traffic volume and draw samples for the entire network. However, recent research indicates that this approach is not the most effective for addressing the variations in the traffic network arising due to differences in roadway geometry and prevailing traffic conditions. Based on the limitations observed in current research and practice, a need for development of a comprehensive sampling design based on statistical sampling theory was identified. A review of sampling design methodology was conducted and stratified sampling was identified as a well suited sampling method for probe based monitoring of a traffic network.;A stratified sampling methodology was developed, implemented and evaluated for probe based monitoring of a traffic network. The key contributions from this research are: (1) First study to develop and implement a rigorous sampling design for probe based traffic monitoring. (2) Stratified Sampling was demonstrated as an appropriate method for application to probe based traffic monitoring. (3) A Genetic Algorithm based design tool was developed as an extension to stratified sampling methodology to generate effective sample allocations for a given strata configuration. (4) For the same range of overall network wide coverage and accuracy, stratified sampling was demonstrated as requiring significantly lower number of samples compared to network wide sampling. |
