Estimating Turning Movements at Signalized Intersections Using Artificial Neural Networks

Abstract

Intersection turning movements' counts are critical input data for traffic studies, analysis, and forecasting. These types of counts are often used to analyze operational performance of signalized intersections under different traffic conditions and peak hours. There are several methods to collect these movements, such as manual counts and video image processing. Traffic studies require turning movement counts although they may not be readily available, especially in the case of traffic forecasting or when seasonal adjustment factors are applied. In these cases, only approach volumes might be available and turning movements are assumed using different techniques to balance the inbound and outbound traffic. Nonetheless, these techniques require initial conditions, and their outcomes are highly sensitive to the initial assumptions. In this paper, a robust and practical method is developed. The turning movement counts at 691 four-leg, and 156 three-leg signalized intersections are analyzed to develop an accurate and reliable turning movement estimation model. A total of 4,175 hours of turning movement counts are used. An artificial neural networks (ANN) model is trained to analyze the relationship between the approach volumes and the corresponding turning movements. The results show that the developed ANN model can be utilized to predict turning movements at a high level of accuracy.