Left-turn vehicles crossing behavior prediction and modeling at mixed traffic flow intersection

Abstract

It is a complicated driving maneuver that left-turn motor vehicles cross conflicted vehicles under non-protected phases, during which drivers' crossing behaviors varies. It is even more complex at mixed-flow intersections where left-turn vehicles interact with both motor and non-motorized vehicles. In this paper, 363 samples of left-turn vehicles crossing motorized straight vehicles, as well as 413 samples of crossing non-motorized straight vehicles are collected. Based on the data, it is found that left-turn vehicles process the entire crossing behaviors in significant two steps at mixed-flow intersections: drivers firstly cross opposite going-straight motor vehicles, and secondly cross non-motorized vehicles. Two logistic models are established to reflect drivers' decision during the two steps of the crossing behaviors. The results show that the prediction accuracies reach 98.9% and 85.2%, respectively. The crossing behaviors are further classified into three types based on their influence on going-straight motor vehicles, which are free crossing, cooperative crossing and forced crossing. Adecision-tree model is built to predict the three crossing behaviors. The predication accuracy is 81.5% on average. The result of this study may help to build a more accurate microscopic simulation model on mixed-flow intersection scenarios.