Accurate travel time estimation has been shown to be crucial information for traffic flow management systems. In this paper, we present the foundations for a travel time estimation methodology using vehicle signatures with special attention to signature pre-processing. This approach to signature pre-processing is based on principled multi-rate signal processing techniques. The matching of signatures from an upstream station to a downstream station is done through a maximum-a-posteriori (MAP) classifier. For each matched upstream-to-downstream pair of signatures, the feasibility of the corresponding travel time is used to filter out mismatches. Travel time is obtained by computing the median of the observations that fall within the modal bin of the travel time histogram. Best results are obtained when only the subset of vehicles with atypical signatures is considered. Although, any types of inductive or magneto-inductive signatures can be appropriately in the presented method, the best performances are expected when the sensors in use generate signatures with a higher degree of pair-wise dissimilarity. Both 3M micro-loops and the standard ILDs (Inductive Loop Detector) were used in this study. In the experimental tests (shorter station separation), the proposed signature pre-processing procedure, when applied to all types of vehicles in the traffic stream and without feasibility constraints, lead to 67.3% re-identification accuracy. When feasibility constraints are still absent and only atypical signatures were considered, the re-identification performance reached 97.4%. These results are very promising since in the absence of feasibility constraints, the performance the re-identification algorithm is independent of the separation between the stations.
Abstract