Estimating roadway horizontal alignment using artificial neural network

Abstract

This paper presents a novel approach for extracting horizontal alignment data from Geographic Information Systems (GIS) centerline shapefiles. Estimating the road horizontal alignment is formulated as a minimization problem, and a two-tiered approach is proposed. Step 1 is the segmentation: determining the curved and tangent sections along a roadway. Step 1 is conducted by applying an artificial neural network (ANN) model, trained using two different datasets, actual and synthetic alignment data, generated using subjective decision on whether a vertex is part of a curved or a tangent section. Step 2 uses the segmentation results and estimates the curvature information using a known algebraic method, called Taubin circle fit. A 10.72 mile long freeway section is used for evaluating the accuracy of the proposed approach, of which the actual alignment information is available. Six different metrics are used for evaluation. The results show the high accuracy of the ANN method, where the overlap of estimated and actual section lengths are 0.97 and 0.92 for curved and tangent sections, respectively.