Automatic Incident Detection for Urban Roads in Hong Kong

Abstract

In this paper, two models are proposed for automatic incident detection on urban roads in Hong Kong. The first model is the threshold-based incident detection model, adopting the ratio of predicted speed over estimated speed as the detection criterion. The second model is established on a basis of the discriminant analysis. These two models are evaluated, in terms of detection rate, false alarm rate and mean time to detect, under different pre-incident traffic conditions (represented as level of service), using the traffic data from a real-time travel information system in Hong Kong. Also, the impacts of level of service on the model detection performance are specified and quantified in this study. The empirical results indicate that the pre-incident traffic conditions (congested or non-congested) may significantly affect the detection performance of models.