Artificial Neural Network–Based Traffic State Estimation Using Erroneous Automated Sensor Data

Abstract

Capturing real-time traffic system characteristics is a primary step in any intelligent transportation system (ITS) application. The majority of the traffic sensors used to capture real-time data have been developed for homogeneous and lane-disciplined traffic conditions. Hence many of them may not perform accurately under heterogeneous and less-lane-disciplined traffic conditions, ultimately leading to reduced estimation accuracy of end applications. The present study addresses this issue by developing an artificial neural network (ANN)–based estimation scheme that can handle these errors and still generate reasonably accurate results. The estimation of location-based speed, stream-based density, and stream speed is carried out using erroneous data as inputs to an ANN trained with accurate data. The same is also performed under varying ranges of errors in inputs. The results show that the ANN can handle the errors in automated data and produce accurate traffic state estimates when trained with good-quality data, hence demonstrating its efficacy for real-time ITS implementation under such traffic conditions.