Transient-based automatic incident detection method for intelligent transport systems

Abstract

One of the major problems of traffic in big cities today is the occurrence of congestion phenomena on the road network, which has several serious effects not only on the lives of drivers, but also on city inhabitants. In order to deal with these phenomena, it is essential to have an in-depth understanding of the processes that lead to the occurrence of congestion and its spilling over into contiguous areas of the city. One of the main causes of congestion phenomena is unexpected traffic incidents on major roads and urban freeways, the rapid and reliable detection of which can help reduce negative impacts. Researching Automatic Incident Detection (AID) has a long history that has again become one of the main subjects of research with the rise of new machine learning methods. Our article presents a new Transient-based Automatic Incident Detection (TBAID) method we have developed, which uses an approach not yet seen in professional literature to detect the occurrence of incidents. The results of our detailed analysis showed that our method performed better than the methods currently available in terms of both speed and reliability on traffic data collected from freeways. We also created a new dataset for the examination of our method, because the datasets used in previous research were either too small or not publicly available. Our dataset contains 452 incidents and data measured with dual loop traffic detectors from the immediate vicinity of incidents, which, to the best of our knowledge, is the largest publicly available incident dataset to date.