Traffic Monitoring and Anomaly Detection based on Simulation of Luxembourg Road Network

Abstract

Traffic incidents which commonly result from traffic accidents, anomalous construction events and inclement weather can cause a wide range of negative impacts on urban road networks. Developing a high efficiency and transferable traffic incident detection system plays an important role in solving the imbalance caused by traffic incidents between traffic demand and capacity. However, the existing literature on transferability of traffic incident detection is rather limited. The objective of this paper is to provide an accurate and transferable incident detection approach based on the relationship between traffic variables and observed traffic incidents, in particular at a network level. We propose a deep learning based method which has been calibrated using part of the collected traffic variables and the pre-assigned traffic incidents and then tested against the rest of the dataset. The proposed method is compared to other benchmarks commonly used in traffic incident detection, in terms of detection rate, false positive rate, f-measurement and detection time. The results indicate that the proposed method is significantly promising for traffic incident detection with high accuracy and transferability compared to the more widely used techniques in the literature.