An Investigation of Intersection Traffic State Classification Using Vehicle Delay Information

The rapid technology evolution in the past decade promises new methods of road traffic data collection with enhanced quantity and quality. These advances are coupled with the development of cost effective platforms to establish sensor and controller networks providing the opportunity to integrate heterogeneous traffic data sources. With these advances and the prospect of big traffic data, it has become possible to measure traffic information such as vehicle travel time/delay information directly. This paper investigates the possibility of using the distribution of vehicle delay information, obtained from historical traffic data, at signalized intersections to classify the traffic state. This classification can be used as a reliable estimate of the traffic demand with various applications such as demand management or traffic signal optimization. The similarities between the delay distributions observed and the traffic state distributions are quantified and compared using several divergence tests including the Chi-Square test, Hellinger distance and the Jensen-Shanon test through simulation. A systematic evaluation under a variety of traffic and driving behavior variable ranges is conducted and the results clearly show that vehicle delay distribution can be used as an effective feature to correctly classify the traffic state using only a few days of historical travel time information and obtain a reliable estimate of the degree of saturation of lane groups at signalized intersections.